TastyPancakes
/
bayesiancnn-data-parsing
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Add code 2023 NVIDIA

This commit is contained in:
Eduardo Cueto-Mendoza 2024-05-10 11:18:47 +01:00
parent e2acd93d18
commit 4783c71929
34 changed files with 16388 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
.gitignore vendored Normal file
View File

@ -0,0 +1,3 @@
*/
*.pkl
*.png

302
100_epoch_exp.jl Executable file
View File

@ -0,0 +1,302 @@
include("aux_func.jl")
using Statistics
using PlotlyJS
using PlotlyJS: savefig
folder = "exp_100_epochs/"
bayes_exp_1 = load_pickle("$(folder)bayes_exp_data_1.pkl")
bayes_exp_2 = load_pickle("$(folder)bayes_exp_data_2.pkl")
bayes_exp_3 = load_pickle("$(folder)bayes_exp_data_3.pkl")
bayes_exp_4 = load_pickle("$(folder)bayes_exp_data_4.pkl")
bayes_exp_5 = load_pickle("$(folder)bayes_exp_data_5.pkl")
bayes_all_mean_tloss =
(1 / 5) * (
bayes_exp_1[:, 2] +
bayes_exp_2[:, 2] +
bayes_exp_3[:, 2] +
bayes_exp_4[:, 2] +
bayes_exp_5[:, 2]
)
bayes_all_mean_iloss =
(1 / 5) * (
bayes_exp_1[:, 4] +
bayes_exp_2[:, 4] +
bayes_exp_3[:, 4] +
bayes_exp_4[:, 4] +
bayes_exp_5[:, 4]
)
bayes_all_mean_acc =
(1 / 5) * (
bayes_exp_1[:, 3] +
bayes_exp_2[:, 3] +
bayes_exp_3[:, 3] +
bayes_exp_4[:, 3] +
bayes_exp_5[:, 3]
)
bayes_all_mean_pre =
(1 / 5) * (
bayes_exp_1[:, 5] +
bayes_exp_2[:, 5] +
bayes_exp_3[:, 5] +
bayes_exp_4[:, 5] +
bayes_exp_5[:, 5]
)
b_exp_1_tls = bayes_exp_1[:, 2]
b_exp_1_acc = bayes_exp_1[:, 3]
b_exp_1_vls = bayes_exp_1[:, 4]
b_exp_1_pre = bayes_exp_1[:, 5]
println("Training accuracy bayes 1 $(mean(b_exp_1_acc))")
println("Testing accuracy bayes 1 $(mean(b_exp_1_pre))")
b_exp_2_tls = bayes_exp_2[:, 2]
b_exp_2_acc = bayes_exp_2[:, 3]
b_exp_2_vls = bayes_exp_2[:, 4]
b_exp_2_pre = bayes_exp_2[:, 5]
println("Training accuracy bayes 2 $(mean(b_exp_2_acc))")
println("Testing accuracy bayes 2 $(mean(b_exp_2_pre))")
b_exp_3_tls = bayes_exp_3[:, 2]
b_exp_3_acc = bayes_exp_3[:, 3]
b_exp_3_vls = bayes_exp_3[:, 4]
b_exp_3_pre = bayes_exp_3[:, 5]
println("Training accuracy bayes 3 $(mean(b_exp_3_acc))")
println("Testing accuracy bayes 3 $(mean(b_exp_3_pre))")
b_exp_4_tls = bayes_exp_4[:, 2]
b_exp_4_acc = bayes_exp_4[:, 3]
b_exp_4_vls = bayes_exp_4[:, 4]
b_exp_4_pre = bayes_exp_4[:, 5]
println("Training accuracy bayes 4 $(mean(b_exp_4_acc))")
println("Testing accuracy bayes 4 $(mean(b_exp_4_pre))")
b_exp_5_tls = bayes_exp_5[:, 2]
b_exp_5_acc = bayes_exp_5[:, 3]
b_exp_5_vls = bayes_exp_5[:, 4]
b_exp_5_pre = bayes_exp_5[:, 5]
println("Training accuracy bayes 5 $(mean(b_exp_5_acc))")
println("Testing accuracy bayes 5 $(mean(b_exp_5_pre))")
cnn_exp_1 = load_pickle("$(folder)freq_exp_data_1.pkl")
cnn_exp_2 = load_pickle("$(folder)freq_exp_data_2.pkl")
cnn_exp_3 = load_pickle("$(folder)freq_exp_data_3.pkl")
cnn_exp_4 = load_pickle("$(folder)freq_exp_data_4.pkl")
cnn_exp_5 = load_pickle("$(folder)freq_exp_data_5.pkl")
cnn_all_mean_tloss =
(1 / 5) * (
cnn_exp_1[:, 2] +
cnn_exp_2[:, 2] +
cnn_exp_3[:, 2] +
cnn_exp_4[:, 2] +
cnn_exp_5[:, 2]
)
cnn_all_mean_iloss =
(1 / 5) * (
cnn_exp_1[:, 4] +
cnn_exp_2[:, 4] +
cnn_exp_3[:, 4] +
cnn_exp_4[:, 4] +
cnn_exp_5[:, 4]
)
cnn_all_mean_acc =
(1 / 5) * (
cnn_exp_1[:, 3] +
cnn_exp_2[:, 3] +
cnn_exp_3[:, 3] +
cnn_exp_4[:, 3] +
cnn_exp_5[:, 3]
)
cnn_all_mean_pre =
(1 / 5) * (
cnn_exp_1[:, 5] +
cnn_exp_2[:, 5] +
cnn_exp_3[:, 5] +
cnn_exp_4[:, 5] +
cnn_exp_5[:, 5]
)
f_exp_1_tls = cnn_exp_1[:, 2]
f_exp_1_acc = cnn_exp_1[:, 3]
f_exp_1_vls = cnn_exp_1[:, 4]
f_exp_1_pre = cnn_exp_1[:, 5]
println("Training accuracy freq 1 $(mean(f_exp_1_acc))")
println("Testing accuracy freq 1 $(mean(f_exp_1_pre))")
f_exp_2_tls = cnn_exp_2[:, 2]
f_exp_2_acc = cnn_exp_2[:, 3]
f_exp_2_vls = cnn_exp_2[:, 4]
f_exp_2_pre = cnn_exp_2[:, 5]
println("Training accuracy freq 2 $(mean(f_exp_2_acc))")
println("Testing accuracy freq 2 $(mean(f_exp_2_pre))")
f_exp_3_tls = cnn_exp_3[:, 2]
f_exp_3_acc = cnn_exp_3[:, 3]
f_exp_3_vls = cnn_exp_3[:, 4]
f_exp_3_pre = cnn_exp_3[:, 5]
println("Training accuracy freq 3 $(mean(f_exp_3_acc))")
println("Testing accuracy freq 3 $(mean(f_exp_3_pre))")
f_exp_4_tls = cnn_exp_4[:, 2]
f_exp_4_acc = cnn_exp_4[:, 3]
f_exp_4_vls = cnn_exp_4[:, 4]
f_exp_4_pre = cnn_exp_4[:, 5]
println("Training accuracy freq 4 $(mean(f_exp_4_acc))")
println("Testing accuracy freq 4 $(mean(f_exp_4_pre))")
f_exp_5_tls = cnn_exp_5[:, 2]
f_exp_5_acc = cnn_exp_5[:, 3]
f_exp_5_vls = cnn_exp_5[:, 4]
f_exp_5_pre = cnn_exp_5[:, 5]
println("Training accuracy freq 5 $(mean(f_exp_5_acc))")
println("Testing accuracy freq 5 $(mean(f_exp_5_pre))")
en_plot = plot(
[
scatter(
y = f_exp_1_acc,
name = "LeNet 1",
marker = attr(color = "rgb(211,120,000)"),
),
scatter(
y = f_exp_2_acc,
name = "LeNet 2",
marker = attr(color = "rgb(255,170,017)"),
),
scatter(
y = f_exp_3_acc,
name = "LeNet 3",
marker = attr(color = "rgb(255,187,034)"),
),
scatter(
y = f_exp_4_acc,
name = "LeNet 4",
marker = attr(color = "rgb(255,204,051)"),
),
scatter(
y = f_exp_5_acc,
name = "LeNet 5",
marker = attr(color = "rgb(255,221,068)"),
),
scatter(
y = b_exp_1_acc,
name = "BCNN 1",
marker = attr(color = "rgb(055,033,240)"),
),
scatter(
y = b_exp_2_acc,
name = "BCNN 2",
marker = attr(color = "rgb(033,081,240)"),
),
scatter(
y = b_exp_3_acc,
name = "BCNN 3",
marker = attr(color = "rgb(033,115,240)"),
),
scatter(
y = b_exp_4_acc,
name = "BCNN 4",
marker = attr(color = "rgb(151,177,255)"),
),
scatter(
y = b_exp_5_acc,
name = "BCNN 5",
marker = attr(color = "rgb(051,215,255)"),
),
],
Layout(
mode = "lines",
opacity = 0.4,
xaxis_tickangle = -45,
yaxis_title_text = "Accuracy",
xaxis_title_text = "Epoch";
yaxis_range = [0, 1],
),
)
savefig(en_plot, "mnist_100_tacc.png")
#=
en_plot = plot([
scatter(y=f_exp_1_pre, name="LeNet 1", marker=attr(color="rgb(211,120,000)")),
scatter(y=f_exp_2_pre, name="LeNet 2", marker=attr(color="rgb(255,170,017)")),
scatter(y=f_exp_3_pre, name="LeNet 3", marker=attr(color="rgb(255,187,034)")),
scatter(y=f_exp_4_pre, name="LeNet 4", marker=attr(color="rgb(255,204,051)")),
scatter(y=f_exp_5_pre, name="LeNet 5", marker=attr(color="rgb(255,221,068)")),
scatter(y=b_exp_1_pre, name="BCNN 1", marker=attr(color="rgb(055,033,240)")),
scatter(y=b_exp_2_pre, name="BCNN 2", marker=attr(color="rgb(033,081,240)")),
scatter(y=b_exp_3_pre, name="BCNN 3", marker=attr(color="rgb(033,115,240)")),
scatter(y=b_exp_4_pre, name="BCNN 4", marker=attr(color="rgb(151,177,255)")),
scatter(y=b_exp_5_pre, name="BCNN 5", marker=attr(color="rgb(051,215,255)"))
], Layout(mode="lines", opacity=0.4, xaxis_tickangle=-45, yaxis_title_text="Accuracy",
xaxis_title_text="Epoch", title="100 Epoch Experiment Testing Accuracy";yaxis_range=[0, 1] ))
savefig(en_plot,"mnist_100_tpre.png")
=#
en_plot = plot(
[
scatter(
y = f_exp_1_pre,
name = "LeNet 1",
marker = attr(color = "rgb(211,120,000)"),
),
scatter(
y = f_exp_2_pre,
name = "LeNet 2",
marker = attr(color = "rgb(255,170,017)"),
),
scatter(
y = f_exp_3_pre,
name = "LeNet 3",
marker = attr(color = "rgb(255,187,034)"),
),
scatter(
y = f_exp_4_pre,
name = "LeNet 4",
marker = attr(color = "rgb(255,204,051)"),
),
scatter(
y = f_exp_5_pre,
name = "LeNet 5",
marker = attr(color = "rgb(255,221,068)"),
),
scatter(
y = b_exp_1_pre,
name = "BCNN 1",
marker = attr(color = "rgb(055,033,240)"),
),
scatter(
y = b_exp_2_pre,
name = "BCNN 2",
marker = attr(color = "rgb(033,081,240)"),
),
scatter(
y = b_exp_3_pre,
name = "BCNN 3",
marker = attr(color = "rgb(033,115,240)"),
),
scatter(
y = b_exp_4_pre,
name = "BCNN 4",
marker = attr(color = "rgb(151,177,255)"),
),
scatter(
y = b_exp_5_pre,
name = "BCNN 5",
marker = attr(color = "rgb(051,215,255)"),
),
],
Layout(
mode = "lines",
opacity = 0.4,
xaxis_tickangle = -45,
yaxis_title_text = "Accuracy",
xaxis_title_text = "Epoch";
yaxis_range = [0, 1],
),
)
savefig(en_plot, "mnist_100_tpre.png")

354
aux_func.jl Executable file
View File

@ -0,0 +1,354 @@
using StatsBase: countmap
using PyCall
function converttobyte(array)
temp = []
for l in array
if l[2] == "Mbyte"
append!(temp, parse(Float64, l[3]) * 1048576.0)
elseif l[2] == "Kbyte"
append!(temp, parse(Float64, l[3]) * 1024.0)
else
append!(temp, parse(Float64, l[3]))
end
end
return temp
end
function approx_δ(array)
temp = []
for i in 1:(length(array)-1)
push!(temp,array[i+1]-array[i])
end
return temp
end
py"""
import pickle
def load_pickle(fpath):
with open(fpath, "rb") as f:
data = pickle.load(f)
return data
"""
load_pickle = py"load_pickle"
py"""
import pickle
def save_pickle(pickle_name, data_dump):
with open(pickle_name, 'wb') as f:
pickle.dump(data_dump, f)
"""
save_pickle = py"save_pickle"
function movavg(X::Vector, numofele::Int)
Back = div(numofele, 2)
Forward = isodd(numofele) ? div(numofele, 2) : div(numofele, 2) - 1
len = length(X)
Y = similar(X)
for n = 1:len
lo = max(1, n - Back)
hi = min(len, n + Forward)
Y[n] = mean(X[lo:hi])
end
return Y
end
function removewatt(X::Vector{String})
temp = match.(r"[0-9]+.[0-9]+|[0-9]+", X)
temp = [x.match for x in temp]
return parse.(Float64, temp)
end
function removewatt(X::SubString{String})
return parse.(Float64, match.(r"[0-9]+.[0-9]+|[0-9]+", X).match)
end
function removewatt(X::String)
return parse(Float64, match(r"[0-9]+.[0-9]+|[0-9]+", X).match)
end
function incwzeros(X::Vector, newsize::Int)
temp = X[:]
l = length(temp)
for i = 1:newsize
push!(temp, 28)
end
return temp
end
function readcpudata(dat_fo, typ, dat_fl, exp)
folder = "$(typ)_$(dat_fo)_cpu_data_$(exp)/"
files_tmp = Dict()
for s = 1:5
file = "$(folder)$(typ)_$(s)_$(dat_fl)"
#println("File being loaded: $(file)")
io = open(file, "r")
t = read(io, String)
close(io)
files_tmp[s] = t
end
return files_tmp
end
function getimes(file_name::String)
io = open(file_name, "r")
times_bayes = read(io, String)
close(io)
times_bayes = split(times_bayes, "\n")
times_bayes = [s for s in times_bayes if !all(isempty.(s))]
times_bayes = [split(r, "|") for r in times_bayes]
temp = Vector()
for vec in times_bayes
t = [strip(s) for s in vec if !all(isempty.(s))]
push!(temp, t)
end
times_bayes = temp
times_bayes = [s for s in times_bayes if s[1] != "------"]
l = length(times_bayes)
times_bayes_mnist = times_bayes[1:(l÷2)]
popfirst!(times_bayes_mnist)
times_bayes_cifar = times_bayes[(l÷2)+1:end]
popfirst!(times_bayes_cifar)
temp = Vector()
for (times, i) in zip(times_bayes_mnist, 1:length(times_bayes_mnist))
if i > 1
t = map(x -> Time(x), times[2:end])
push!(temp, t)
end
end
times_bayes_mnist = temp
temp = Vector()
for (times, i) in zip(times_bayes_cifar, 1:length(times_bayes_cifar))
if i > 1
t = map(x -> Time(x), times[2:end])
push!(temp, t)
end
end
times_bayes_cifar = temp
temp = Vector()
for (p, o) in zip(2:2:10, 1:2:10)
t = convert.(Dates.Second, times_bayes_mnist[p] .- times_bayes_mnist[o])
push!(temp, t)
end
times_bayes_mnist = temp
temp = Vector()
for (p, o) in zip(2:2:10, 1:2:10)
t = convert.(Dates.Second, times_bayes_cifar[p] .- times_bayes_cifar[o])
push!(temp, t)
end
times_bayes_cifar = temp
if file_name == "times_frequentist"
typ = "Frequentist"
elseif file_name == "times_bayesian"
typ = "Bayesian"
end
to_file = Vector()
push!(to_file, "\nMNIST $(typ) Results: \n")
push!(to_file, " acc > 99 ES W Budget 100 Epochs \n")
for i = 1:5
push!(
to_file,
"size $(i): $(times_bayes_mnist[i][1]) $(times_bayes_mnist[i][2]) $(times_bayes_mnist[i][3]) $(times_bayes_mnist[i][4]) \n",
)
end
push!(to_file, "\nCIFAR $(typ) Results: \n")
push!(to_file, " acc > 99 ES W Budget 100 Epochs\n")
for i = 1:5
push!(
to_file,
"size $(i): $(times_bayes_cifar[i][1]) $(times_bayes_cifar[i][2]) $(times_bayes_cifar[i][3]) $(times_bayes_cifar[i][4]) \n",
)
end
return to_file
end
function getcpuwatt(watt_vector)
test = watt_vector
test = split(test, "\n")
test = [s for s in test if !all(isempty.(s))]
test = [split(r, " ") for r in test]
temp = Vector()
for vec in test
if occursin(r"\d\d:\d\d:\d\d", vec[1])
push!(temp, vec)
end
end
test = temp
temp = Vector()
for vec in test
t = [strip(s) for s in vec if !all(isempty.(s))]
push!(temp, t)
end
test = temp
temp = Vector()
for vec in test
push!(temp, vec[end])
end
test = temp
test = map(x -> parse(Float32, x), test)
return test
end
function getramuse(ram_vector)
test = ram_vector
test = split(test, "\n")
test = [s for s in test if !all(isempty.(s))]
test = [split(r, " ") for r in test]
temp = Vector()
for vec in test
t = [strip(s) for s in vec if !all(isempty.(s))]
push!(temp, t)
end
test = temp
temp = Vector()
for vec in test
if vec[1] == "Mem:"
push!(temp, vec)
end
end
test = temp
temp = Vector()
regex = r"[^0-9]+$"
for vec in test
unit = match(regex, vec[3]).match
qty = removewatt(vec[3])
push!(temp, (qty, unit))
end
test = temp
temp = Vector()
for u in test
if u[2] == "Gi"
push!(temp, u[1] * 1024.0)
elseif u[2] == "Mi"
println(u[1])
end
end
test = temp
return test
end
function concat(arr::Vector)
temp = ""
for s in arr
temp = temp * s
end
return temp
end
function getgpudata(folder, model, type, save = "d")
temp = Dict()
for s = 1:5
if type == "watt"
path = "$(folder)$(model)_$(type)data_$(s).pkl"
watt_data = removewatt(load_pickle(path)[:, 1])
mem_data = removewatt(load_pickle(path)[:, 2])
if save == "d"
temp[s] = Dict("Ene" => watt_data, "Mem" => mem_data)
elseif save == "eo"
temp[s] = Dict("Ene" => watt_data, "Mem" => nothing)
elseif save == "mo"
temp[s] = Dict("Ene" => nothing, "Mem" => mem_data)
end
elseif type == "exp"
path = "$(folder)$(model)_$(type)_data_$(s).pkl"
temp[s] = load_pickle(path)
end
end
return temp
end
function getuniquevalues(vector::Vector)::Tuple
dict = countmap(vector)
uniq = unique(vector)
vals = Vector()
for u in uniq
push!(vals, dict[u])
end
return (uniq, vals)
end
function expdatatodict(matrix::Matrix{Real})::Dict
return Dict(
"tloss" => matrix[:, 2],
"acc" => matrix[:, 3],
"vloss" => matrix[:, 4],
"pre" => matrix[:, 5],
)
end
@views function makechunks(X::AbstractVector, n::Integer)
c = length(X) ÷ n
return [X[1+c*k:(k == n-1 ? end : c*k+c)] for k = 0:n-1]
end
function resize1(vect)
temp = []
size = length(vect[end])
for v in vect
temp_size = length(v)
if temp_size == size
push!(temp, Array(v))
elseif temp_size < size
app = zeros(size - temp_size)
x = Array(deepcopy(v))
new = vcat(x,app)
push!(temp, new)
elseif temp_size > size
resize!(Array(v),size)
push!(temp, v)
end
end
return temp
end
function join_vectors(vect)
result = []
temp = []
for v in vect
temp = vcat(temp,v)
push!(result,temp)
end
return result
end

238
bounded.jl Executable file
View File

@ -0,0 +1,238 @@
####################################################
#
# Accuracy is bounded
#
#####################################################
include("aux_func.jl")
using PlotlyJS
using PlotlyJS: savefig
using Statistics
folder = "data_bounded/"
bayes_exp_1 = load_pickle("$(folder)bayes_exp_data_1.pkl");
bayes_exp_2 = load_pickle("$(folder)bayes_exp_data_2.pkl");
bayes_exp_3 = load_pickle("$(folder)bayes_exp_data_3.pkl");
bayes_exp_4 = load_pickle("$(folder)bayes_exp_data_4.pkl");
bayes_exp_5 = load_pickle("$(folder)bayes_exp_data_5.pkl");
b_exp_1_tls = bayes_exp_1[:, 2]
b_exp_1_acc = bayes_exp_1[:, 3]
b_exp_1_vls = bayes_exp_1[:, 4]
b_exp_1_pre = bayes_exp_1[:, 5]
println("Training accuracy bayes 1 $(mean(b_exp_1_acc))")
println("Testing accuracy bayes 1 $(mean(b_exp_1_pre))")
b_exp_2_tls = bayes_exp_2[:, 2]
b_exp_2_acc = bayes_exp_2[:, 3]
b_exp_2_vls = bayes_exp_2[:, 4]
b_exp_2_pre = bayes_exp_2[:, 5]
println("Training accuracy bayes 2 $(mean(b_exp_2_acc))")
println("Testing accuracy bayes 2 $(mean(b_exp_2_pre))")
b_exp_3_tls = bayes_exp_3[:, 2]
b_exp_3_acc = bayes_exp_3[:, 3]
b_exp_3_vls = bayes_exp_3[:, 4]
b_exp_3_pre = bayes_exp_3[:, 5]
println("Training accuracy bayes 3 $(mean(b_exp_3_acc))")
println("Testing accuracy bayes 3 $(mean(b_exp_3_pre))")
b_exp_4_tls = bayes_exp_4[:, 2]
b_exp_4_acc = bayes_exp_4[:, 3]
b_exp_4_vls = bayes_exp_4[:, 4]
b_exp_4_pre = bayes_exp_4[:, 5]
println("Training accuracy bayes 4 $(mean(b_exp_4_acc))")
println("Testing accuracy bayes 4 $(mean(b_exp_4_pre))")
b_exp_5_tls = bayes_exp_5[:, 2]
b_exp_5_acc = bayes_exp_5[:, 3]
b_exp_5_vls = bayes_exp_5[:, 4]
b_exp_5_pre = bayes_exp_5[:, 5]
println("Training accuracy bayes 5 $(mean(b_exp_5_acc))")
println("Testing accuracy bayes 5 $(mean(b_exp_5_pre))")
println("Amount of epochs to reach bound Bayesian 1x: $(size(bayes_exp_1)[1])")
println("Amount of epochs to reach bound Bayesian 2x: $(size(bayes_exp_2)[1])")
println("Amount of epochs to reach bound Bayesian 3x: $(size(bayes_exp_3)[1])")
println("Amount of epochs to reach bound Bayesian 4x: $(size(bayes_exp_4)[1])")
println("Amount of epochs to reach bound Bayesian 5x: $(size(bayes_exp_5)[1])")
freq_exp_1 = load_pickle("$(folder)freq_exp_data_1.pkl");
freq_exp_2 = load_pickle("$(folder)freq_exp_data_2.pkl");
freq_exp_3 = load_pickle("$(folder)freq_exp_data_3.pkl");
freq_exp_4 = load_pickle("$(folder)freq_exp_data_4.pkl");
freq_exp_5 = load_pickle("$(folder)freq_exp_data_5.pkl");
f_exp_1_tls = freq_exp_1[:, 2]
f_exp_1_acc = freq_exp_1[:, 3]
f_exp_1_vls = freq_exp_1[:, 4]
f_exp_1_pre = freq_exp_1[:, 5]
println("Training accuracy freq 1 $(mean(f_exp_1_acc))")
println("Testing accuracy freq 1 $(mean(f_exp_1_pre))")
f_exp_2_tls = freq_exp_2[:, 2]
f_exp_2_acc = freq_exp_2[:, 3]
f_exp_2_vls = freq_exp_2[:, 4]
f_exp_2_pre = freq_exp_2[:, 5]
println("Training accuracy freq 2 $(mean(f_exp_2_acc))")
println("Testing accuracy freq 2 $(mean(f_exp_2_pre))")
f_exp_3_tls = freq_exp_3[:, 2]
f_exp_3_acc = freq_exp_3[:, 3]
f_exp_3_vls = freq_exp_3[:, 4]
f_exp_3_pre = freq_exp_3[:, 5]
println("Training accuracy freq 3 $(mean(f_exp_3_acc))")
println("Testing accuracy freq 3 $(mean(f_exp_3_pre))")
f_exp_4_tls = freq_exp_4[:, 2]
f_exp_4_acc = freq_exp_4[:, 3]
f_exp_4_vls = freq_exp_4[:, 4]
f_exp_4_pre = freq_exp_4[:, 5]
println("Training accuracy freq 4 $(mean(f_exp_4_acc))")
println("Testing accuracy freq 4 $(mean(f_exp_4_pre))")
f_exp_5_tls = freq_exp_5[:, 2]
f_exp_5_acc = freq_exp_5[:, 3]
f_exp_5_vls = freq_exp_5[:, 4]
f_exp_5_pre = freq_exp_5[:, 5]
println("Training accuracy freq 5 $(mean(f_exp_5_acc))")
println("Testing accuracy freq 5 $(mean(f_exp_5_pre))")
println("Amount of epochs to reach bound LeNet 1x: $(size(freq_exp_1)[1])")
println("Amount of epochs to reach bound LeNet 2x: $(size(freq_exp_2)[1])")
println("Amount of epochs to reach bound LeNet 3x: $(size(freq_exp_3)[1])")
println("Amount of epochs to reach bound LeNet 4x: $(size(freq_exp_4)[1])")
println("Amount of epochs to reach bound LeNet 5x: $(size(freq_exp_5)[1])")
en_plot = plot(
[
scatter(
y = f_exp_1_acc,
name = "LeNet 1",
marker = attr(color = "rgb(211,120,000)"),
),
scatter(
y = f_exp_2_acc,
name = "LeNet 2",
marker = attr(color = "rgb(255,170,017)"),
),
scatter(
y = f_exp_3_acc,
name = "LeNet 3",
marker = attr(color = "rgb(255,187,034)"),
),
scatter(
y = f_exp_4_acc,
name = "LeNet 4",
marker = attr(color = "rgb(255,204,051)"),
),
scatter(
y = f_exp_5_acc,
name = "LeNet 5",
marker = attr(color = "rgb(255,221,068)"),
),
scatter(
y = b_exp_1_acc,
name = "BCNN 1",
marker = attr(color = "rgb(055,033,240)"),
),
scatter(
y = b_exp_2_acc,
name = "BCNN 2",
marker = attr(color = "rgb(033,081,240)"),
),
scatter(
y = b_exp_3_acc,
name = "BCNN 3",
marker = attr(color = "rgb(033,115,240)"),
),
scatter(
y = b_exp_4_acc,
name = "BCNN 4",
marker = attr(color = "rgb(151,177,255)"),
),
scatter(
y = b_exp_5_acc,
name = "BCNN 5",
marker = attr(color = "rgb(051,215,255)"),
),
],
Layout(
mode = "lines",
opacity = 0.4,
xaxis_tickangle = -45,
yaxis_title_text = "Accuracy",
xaxis_title_text = "Epoch";
yaxis_range = [0, 1],
),
)
savefig(en_plot, "mnist_ab_tacc.png")
en_plot = plot(
[
scatter(
y = f_exp_1_pre,
name = "LeNet 1",
marker = attr(color = "rgb(211,120,000)"),
),
scatter(
y = f_exp_2_pre,
name = "LeNet 2",
marker = attr(color = "rgb(255,170,017)"),
),
scatter(
y = f_exp_3_pre,
name = "LeNet 3",
marker = attr(color = "rgb(255,187,034)"),
),
scatter(
y = f_exp_4_pre,
name = "LeNet 4",
marker = attr(color = "rgb(255,204,051)"),
),
scatter(
y = f_exp_5_pre,
name = "LeNet 5",
marker = attr(color = "rgb(255,221,068)"),
),
scatter(
y = b_exp_1_pre,
name = "BCNN 1",
marker = attr(color = "rgb(055,033,240)"),
),
scatter(
y = b_exp_2_pre,
name = "BCNN 2",
marker = attr(color = "rgb(033,081,240)"),
),
scatter(
y = b_exp_3_pre,
name = "BCNN 3",
marker = attr(color = "rgb(033,115,240)"),
),
scatter(
y = b_exp_4_pre,
name = "BCNN 4",
marker = attr(color = "rgb(151,177,255)"),
),
scatter(
y = b_exp_5_pre,
name = "BCNN 5",
marker = attr(color = "rgb(051,215,255)"),
),
],
Layout(
mode = "lines",
opacity = 0.4,
xaxis_tickangle = -45,
yaxis_title_text = "Accuracy",
xaxis_title_text = "Epoch";
yaxis_range = [0, 1],
),
)
savefig(en_plot, "mnist_ab_tpre.png")

236
budget.jl Executable file
View File

@ -0,0 +1,236 @@
####################################################
#
# Energy is bounded
#
#####################################################
include("aux_func.jl")
using Statistics
using PlotlyJS
using PlotlyJS: savefig
folder = "data_budget/"
bayes_exp_1 = load_pickle("$(folder)bayes_exp_data_1.pkl");
bayes_exp_2 = load_pickle("$(folder)bayes_exp_data_2.pkl");
bayes_exp_3 = load_pickle("$(folder)bayes_exp_data_3.pkl");
bayes_exp_4 = load_pickle("$(folder)bayes_exp_data_4.pkl");
bayes_exp_5 = load_pickle("$(folder)bayes_exp_data_5.pkl");
b_exp_1_tls = bayes_exp_1[:, 2]
b_exp_1_acc = bayes_exp_1[:, 3]
b_exp_1_vls = bayes_exp_1[:, 4]
b_exp_1_pre = bayes_exp_1[:, 5]
println("Training accuracy bayes 1 $(mean(b_exp_1_acc))")
println("Testing accuracy bayes 1 $(mean(b_exp_1_pre))")
b_exp_2_tls = bayes_exp_2[:, 2]
b_exp_2_acc = bayes_exp_2[:, 3]
b_exp_2_vls = bayes_exp_2[:, 4]
b_exp_2_pre = bayes_exp_2[:, 5]
println("Training accuracy bayes 2 $(mean(b_exp_2_acc))")
println("Testing accuracy bayes 2 $(mean(b_exp_2_pre))")
b_exp_3_tls = bayes_exp_3[:, 2]
b_exp_3_acc = bayes_exp_3[:, 3]
b_exp_3_vls = bayes_exp_3[:, 4]
b_exp_3_pre = bayes_exp_3[:, 5]
println("Training accuracy bayes 3 $(mean(b_exp_3_acc))")
println("Testing accuracy bayes 3 $(mean(b_exp_3_pre))")
b_exp_4_tls = bayes_exp_4[:, 2]
b_exp_4_acc = bayes_exp_4[:, 3]
b_exp_4_vls = bayes_exp_4[:, 4]
b_exp_4_pre = bayes_exp_4[:, 5]
println("Training accuracy bayes 4 $(mean(b_exp_4_acc))")
println("Testing accuracy bayes 4 $(mean(b_exp_4_pre))")
b_exp_5_tls = bayes_exp_5[:, 2]
b_exp_5_acc = bayes_exp_5[:, 3]
b_exp_5_vls = bayes_exp_5[:, 4]
b_exp_5_pre = bayes_exp_5[:, 5]
println("Training accuracy bayes 5 $(mean(b_exp_5_acc))")
println("Testing accuracy bayes 5 $(mean(b_exp_5_pre))")
println("Amount of epochs to reach bound Bayesian 1x: $(size(bayes_exp_1)[1])")
println("Amount of epochs to reach bound Bayesian 2x: $(size(bayes_exp_2)[1])")
println("Amount of epochs to reach bound Bayesian 3x: $(size(bayes_exp_3)[1])")
println("Amount of epochs to reach bound Bayesian 4x: $(size(bayes_exp_4)[1])")
println("Amount of epochs to reach bound Bayesian 5x: $(size(bayes_exp_5)[1])")
freq_exp_1 = load_pickle("$(folder)freq_exp_data_1.pkl");
freq_exp_2 = load_pickle("$(folder)freq_exp_data_2.pkl");
freq_exp_3 = load_pickle("$(folder)freq_exp_data_3.pkl");
freq_exp_4 = load_pickle("$(folder)freq_exp_data_4.pkl");
freq_exp_5 = load_pickle("$(folder)freq_exp_data_5.pkl");
f_exp_1_tls = freq_exp_1[:, 2]
f_exp_1_acc = freq_exp_1[:, 3]
f_exp_1_vls = freq_exp_1[:, 4]
f_exp_1_pre = freq_exp_1[:, 5]
println("Training accuracy freq 1 $(mean(f_exp_1_acc))")
println("Testing accuracy freq 1 $(mean(f_exp_1_pre))")
f_exp_2_tls = freq_exp_2[:, 2]
f_exp_2_acc = freq_exp_2[:, 3]
f_exp_2_vls = freq_exp_2[:, 4]
f_exp_2_pre = freq_exp_2[:, 5]
println("Training accuracy freq 2 $(mean(f_exp_2_acc))")
println("Testing accuracy freq 2 $(mean(f_exp_2_pre))")
f_exp_3_tls = freq_exp_3[:, 2]
f_exp_3_acc = freq_exp_3[:, 3]
f_exp_3_vls = freq_exp_3[:, 4]
f_exp_3_pre = freq_exp_3[:, 5]
println("Training accuracy freq 3 $(mean(f_exp_3_acc))")
println("Testing accuracy freq 3 $(mean(f_exp_3_pre))")
f_exp_4_tls = freq_exp_4[:, 2]
f_exp_4_acc = freq_exp_4[:, 3]
f_exp_4_vls = freq_exp_4[:, 4]
f_exp_4_pre = freq_exp_4[:, 5]
println("Training accuracy freq 4 $(mean(f_exp_4_acc))")
println("Testing accuracy freq 4 $(mean(f_exp_4_pre))")
f_exp_5_tls = freq_exp_5[:, 2]
f_exp_5_acc = freq_exp_5[:, 3]
f_exp_5_vls = freq_exp_5[:, 4]
f_exp_5_pre = freq_exp_5[:, 5]
println("Training accuracy freq 5 $(mean(f_exp_5_acc))")
println("Testing accuracy freq 5 $(mean(f_exp_5_pre))")
println("Amount of epochs to reach bound LeNet 1x: $(size(freq_exp_1)[1])")
println("Amount of epochs to reach bound LeNet 2x: $(size(freq_exp_2)[1])")
println("Amount of epochs to reach bound LeNet 3x: $(size(freq_exp_3)[1])")
println("Amount of epochs to reach bound LeNet 4x: $(size(freq_exp_4)[1])")
println("Amount of epochs to reach bound LeNet 5x: $(size(freq_exp_5)[1])")
en_plot = plot(
[
scatter(
y = f_exp_1_acc,
name = "LeNet 1",
marker = attr(color = "rgb(211,120,000)"),
),
scatter(
y = f_exp_2_acc,
name = "LeNet 2",
marker = attr(color = "rgb(255,170,017)"),
),
scatter(
y = f_exp_3_acc,
name = "LeNet 3",
marker = attr(color = "rgb(255,187,034)"),
),
scatter(
y = f_exp_4_acc,
name = "LeNet 4",
marker = attr(color = "rgb(255,204,051)"),
),
scatter(
y = f_exp_5_acc,
name = "LeNet 5",
marker = attr(color = "rgb(255,221,068)"),
),
scatter(
y = b_exp_1_acc,
name = "BCNN 1",
marker = attr(color = "rgb(055,033,240)"),
),
scatter(
y = b_exp_2_acc,
name = "BCNN 2",
marker = attr(color = "rgb(033,081,240)"),
),
scatter(
y = b_exp_3_acc,
name = "BCNN 3",
marker = attr(color = "rgb(033,115,240)"),
),
scatter(
y = b_exp_4_acc,
name = "BCNN 4",
marker = attr(color = "rgb(151,177,255)"),
),
scatter(
y = b_exp_5_acc,
name = "BCNN 5",
marker = attr(color = "rgb(051,215,255)"),
),
],
Layout(
mode = "lines",
opacity = 0.4,
xaxis_tickangle = -45,
yaxis_title_text = "Accuracy",
xaxis_title_text = "Epoch";
yaxis_range = [0, 1],
),
)
savefig(en_plot, "mnist_eb_tacc.png")
en_plot = plot(
[
scatter(
y = f_exp_1_pre,
name = "LeNet 1",
marker = attr(color = "rgb(211,120,000)"),
),
scatter(
y = f_exp_2_pre,
name = "LeNet 2",
marker = attr(color = "rgb(255,170,017)"),
),
scatter(
y = f_exp_3_pre,
name = "LeNet 3",
marker = attr(color = "rgb(255,187,034)"),
),
scatter(
y = f_exp_4_pre,
name = "LeNet 4",
marker = attr(color = "rgb(255,204,051)"),
),
scatter(
y = f_exp_5_pre,
name = "LeNet 5",
marker = attr(color = "rgb(255,221,068)"),
),
scatter(
y = b_exp_1_pre,
name = "BCNN 1",
marker = attr(color = "rgb(055,033,240)"),
),
scatter(
y = b_exp_2_pre,
name = "BCNN 2",
marker = attr(color = "rgb(033,081,240)"),
),
scatter(
y = b_exp_3_pre,
name = "BCNN 3",
marker = attr(color = "rgb(033,115,240)"),
),
scatter(
y = b_exp_4_pre,
name = "BCNN 4",
marker = attr(color = "rgb(151,177,255)"),
),
scatter(
y = b_exp_5_pre,
name = "BCNN 5",
marker = attr(color = "rgb(051,215,255)"),
),
],
Layout(
mode = "lines",
opacity = 0.4,
xaxis_tickangle = -45,
yaxis_title_text = "Accuracy",
xaxis_title_text = "Epoch";
yaxis_range = [0, 1],
),
)
savefig(en_plot, "mnist_eb_tpre.png")

293
cifar_100_epoch.jl Executable file
View File

@ -0,0 +1,293 @@
include("aux_func.jl")
using Statistics
using PlotlyJS
using PlotlyJS: savefig
folder = "CIFAR_100_epoch/"
bayes_exp_1 = load_pickle("$(folder)bayes_exp_data_1.pkl")
bayes_exp_2 = load_pickle("$(folder)bayes_exp_data_2.pkl")
bayes_exp_3 = load_pickle("$(folder)bayes_exp_data_3.pkl")
bayes_exp_4 = load_pickle("$(folder)bayes_exp_data_4.pkl")
bayes_exp_5 = load_pickle("$(folder)bayes_exp_data_5.pkl")
bayes_all_mean_tloss =
(1 / 5) * (
bayes_exp_1[:, 2] +
bayes_exp_2[:, 2] +
bayes_exp_3[:, 2] +
bayes_exp_4[:, 2] +
bayes_exp_5[:, 2]
)
bayes_all_mean_iloss =
(1 / 5) * (
bayes_exp_1[:, 4] +
bayes_exp_2[:, 4] +
bayes_exp_3[:, 4] +
bayes_exp_4[:, 4] +
bayes_exp_5[:, 4]
)
bayes_all_mean_acc =
(1 / 5) * (
bayes_exp_1[:, 3] +
bayes_exp_2[:, 3] +
bayes_exp_3[:, 3] +
bayes_exp_4[:, 3] +
bayes_exp_5[:, 3]
)
bayes_all_mean_pre =
(1 / 5) * (
bayes_exp_1[:, 5] +
bayes_exp_2[:, 5] +
bayes_exp_3[:, 5] +
bayes_exp_4[:, 5] +
bayes_exp_5[:, 5]
)
b_exp_1_tls = bayes_exp_1[:, 2]
b_exp_1_acc = bayes_exp_1[:, 3]
b_exp_1_vls = bayes_exp_1[:, 4]
b_exp_1_pre = bayes_exp_1[:, 5]
println("Training accuracy bayes 1 $(mean(b_exp_1_acc))")
println("Testing accuracy bayes 1 $(mean(b_exp_1_pre))")
b_exp_2_tls = bayes_exp_2[:, 2]
b_exp_2_acc = bayes_exp_2[:, 3]
b_exp_2_vls = bayes_exp_2[:, 4]
b_exp_2_pre = bayes_exp_2[:, 5]
println("Training accuracy bayes 2 $(mean(b_exp_2_acc))")
println("Testing accuracy bayes 2 $(mean(b_exp_2_pre))")
b_exp_3_tls = bayes_exp_3[:, 2]
b_exp_3_acc = bayes_exp_3[:, 3]
b_exp_3_vls = bayes_exp_3[:, 4]
b_exp_3_pre = bayes_exp_3[:, 5]
println("Training accuracy bayes 3 $(mean(b_exp_3_acc))")
println("Testing accuracy bayes 3 $(mean(b_exp_3_pre))")
b_exp_4_tls = bayes_exp_4[:, 2]
b_exp_4_acc = bayes_exp_4[:, 3]
b_exp_4_vls = bayes_exp_4[:, 4]
b_exp_4_pre = bayes_exp_4[:, 5]
println("Training accuracy bayes 4 $(mean(b_exp_4_acc))")
println("Testing accuracy bayes 4 $(mean(b_exp_4_pre))")
b_exp_5_tls = bayes_exp_5[:, 2]
b_exp_5_acc = bayes_exp_5[:, 3]
b_exp_5_vls = bayes_exp_5[:, 4]
b_exp_5_pre = bayes_exp_5[:, 5]
println("Training accuracy bayes 5 $(mean(b_exp_5_acc))")
println("Testing accuracy bayes 5 $(mean(b_exp_5_pre))")
cnn_exp_1 = load_pickle("$(folder)freq_exp_data_1.pkl")
cnn_exp_2 = load_pickle("$(folder)freq_exp_data_2.pkl")
cnn_exp_3 = load_pickle("$(folder)freq_exp_data_3.pkl")
cnn_exp_4 = load_pickle("$(folder)freq_exp_data_4.pkl")
cnn_exp_5 = load_pickle("$(folder)freq_exp_data_5.pkl")
cnn_all_mean_tloss =
(1 / 5) * (
cnn_exp_1[:, 2] +
cnn_exp_2[:, 2] +
cnn_exp_3[:, 2] +
cnn_exp_4[:, 2] +
cnn_exp_5[:, 2]
)
cnn_all_mean_iloss =
(1 / 5) * (
cnn_exp_1[:, 4] +
cnn_exp_2[:, 4] +
cnn_exp_3[:, 4] +
cnn_exp_4[:, 4] +
cnn_exp_5[:, 4]
)
cnn_all_mean_acc =
(1 / 5) * (
cnn_exp_1[:, 3] +
cnn_exp_2[:, 3] +
cnn_exp_3[:, 3] +
cnn_exp_4[:, 3] +
cnn_exp_5[:, 3]
)
cnn_all_mean_pre =
(1 / 5) * (
cnn_exp_1[:, 5] +
cnn_exp_2[:, 5] +
cnn_exp_3[:, 5] +
cnn_exp_4[:, 5] +
cnn_exp_5[:, 5]
)
f_exp_1_tls = cnn_exp_1[:, 2]
f_exp_1_acc = cnn_exp_1[:, 3]
f_exp_1_vls = cnn_exp_1[:, 4]
f_exp_1_pre = cnn_exp_1[:, 5]
println("Training accuracy freq 1 $(mean(f_exp_1_acc))")
println("Testing accuracy freq 1 $(mean(f_exp_1_pre))")
f_exp_2_tls = cnn_exp_2[:, 2]
f_exp_2_acc = cnn_exp_2[:, 3]
f_exp_2_vls = cnn_exp_2[:, 4]
f_exp_2_pre = cnn_exp_2[:, 5]
println("Training accuracy freq 2 $(mean(f_exp_2_acc))")
println("Testing accuracy freq 2 $(mean(f_exp_2_pre))")
f_exp_3_tls = cnn_exp_3[:, 2]
f_exp_3_acc = cnn_exp_3[:, 3]
f_exp_3_vls = cnn_exp_3[:, 4]
f_exp_3_pre = cnn_exp_3[:, 5]
println("Training accuracy freq 3 $(mean(f_exp_3_acc))")
println("Testing accuracy freq 3 $(mean(f_exp_3_pre))")
f_exp_4_tls = cnn_exp_4[:, 2]
f_exp_4_acc = cnn_exp_4[:, 3]
f_exp_4_vls = cnn_exp_4[:, 4]
f_exp_4_pre = cnn_exp_4[:, 5]
println("Training accuracy freq 4 $(mean(f_exp_4_acc))")
println("Testing accuracy freq 4 $(mean(f_exp_4_pre))")
f_exp_5_tls = cnn_exp_5[:, 2]
f_exp_5_acc = cnn_exp_5[:, 3]
f_exp_5_vls = cnn_exp_5[:, 4]
f_exp_5_pre = cnn_exp_5[:, 5]
println("Training accuracy freq 5 $(mean(f_exp_5_acc))")
println("Testing accuracy freq 5 $(mean(f_exp_5_pre))")
en_plot = plot(
[
scatter(
y = f_exp_1_acc,
name = "LeNet 1",
marker = attr(color = "rgb(211,120,000)"),
),
scatter(
y = f_exp_2_acc,
name = "LeNet 2",
marker = attr(color = "rgb(255,170,017)"),
),
scatter(
y = f_exp_3_acc,
name = "LeNet 3",
marker = attr(color = "rgb(255,187,034)"),
),
scatter(
y = f_exp_4_acc,
name = "LeNet 4",
marker = attr(color = "rgb(255,204,051)"),
),
scatter(
y = f_exp_5_acc,
name = "LeNet 5",
marker = attr(color = "rgb(255,221,068)"),
),
scatter(
y = b_exp_1_acc,
name = "BCNN 1",
marker = attr(color = "rgb(055,033,240)"),
),
scatter(
y = b_exp_2_acc,
name = "BCNN 2",
marker = attr(color = "rgb(033,081,240)"),
),
scatter(
y = b_exp_3_acc,
name = "BCNN 3",
marker = attr(color = "rgb(033,115,240)"),
),
scatter(
y = b_exp_4_acc,
name = "BCNN 4",
marker = attr(color = "rgb(151,177,255)"),
),
scatter(
y = b_exp_5_acc,
name = "BCNN 5",
marker = attr(color = "rgb(051,215,255)"),
),
],
Layout(
mode = "lines",
opacity = 0.4,
xaxis_tickangle = -45,
yaxis_title_text = "Accuracy",
xaxis_title_text = "Epoch";
yaxis_range = [0, 1],
),
)
savefig(en_plot, "cifar_100_tacc.png")
en_plot = plot(
[
scatter(
y = f_exp_1_pre,
name = "LeNet 1",
marker = attr(color = "rgb(211,120,000)"),
),
scatter(
y = f_exp_2_pre,
name = "LeNet 2",
marker = attr(color = "rgb(255,170,017)"),
),
scatter(
y = f_exp_3_pre,
name = "LeNet 3",
marker = attr(color = "rgb(255,187,034)"),
),
scatter(
y = f_exp_4_pre,
name = "LeNet 4",
marker = attr(color = "rgb(255,204,051)"),
),
scatter(
y = f_exp_5_pre,
name = "LeNet 5",
marker = attr(color = "rgb(255,221,068)"),
),
scatter(
y = b_exp_1_pre,
name = "BCNN 1",
marker = attr(color = "rgb(055,033,240)"),
),
scatter(
y = b_exp_2_pre,
name = "BCNN 2",
marker = attr(color = "rgb(033,081,240)"),
),
scatter(
y = b_exp_3_pre,
name = "BCNN 3",
marker = attr(color = "rgb(033,115,240)"),
),
scatter(
y = b_exp_4_pre,
name = "BCNN 4",
marker = attr(color = "rgb(151,177,255)"),
),
scatter(
y = b_exp_5_pre,
name = "BCNN 5",
marker = attr(color = "rgb(051,215,255)"),
),
],
Layout(
mode = "lines",
opacity = 0.4,
xaxis_tickangle = -45,
yaxis_title_text = "Accuracy",
xaxis_title_text = "Epoch";
yaxis_range = [0, 1],
),
)
savefig(en_plot, "cifar_100_tpre.png")

236
cifar_bounded.jl Executable file
View File

@ -0,0 +1,236 @@
####################################################
#
# Accuracy is bounded
#
#####################################################
include("aux_func.jl")
using Statistics
using PlotlyJS
using PlotlyJS: savefig
folder = "CIFAR_acc_bound/"
bayes_exp_1 = load_pickle("$(folder)bayes_exp_data_1.pkl");
bayes_exp_2 = load_pickle("$(folder)bayes_exp_data_2.pkl");
bayes_exp_3 = load_pickle("$(folder)bayes_exp_data_3.pkl");
bayes_exp_4 = load_pickle("$(folder)bayes_exp_data_4.pkl");
bayes_exp_5 = load_pickle("$(folder)bayes_exp_data_5.pkl");
b_exp_1_tls = bayes_exp_1[:, 2]
b_exp_1_acc = bayes_exp_1[:, 3]
b_exp_1_vls = bayes_exp_1[:, 4]
b_exp_1_pre = bayes_exp_1[:, 5]
println("Training accuracy bayes 1 $(mean(b_exp_1_acc))")
println("Testing accuracy bayes 1 $(mean(b_exp_1_pre))")
b_exp_2_tls = bayes_exp_2[:, 2]
b_exp_2_acc = bayes_exp_2[:, 3]
b_exp_2_vls = bayes_exp_2[:, 4]
b_exp_2_pre = bayes_exp_2[:, 5]
println("Training accuracy bayes 2 $(mean(b_exp_2_acc))")
println("Testing accuracy bayes 2 $(mean(b_exp_2_pre))")
b_exp_3_tls = bayes_exp_3[:, 2]
b_exp_3_acc = bayes_exp_3[:, 3]
b_exp_3_vls = bayes_exp_3[:, 4]
b_exp_3_pre = bayes_exp_3[:, 5]
println("Training accuracy bayes 3 $(mean(b_exp_3_acc))")
println("Testing accuracy bayes 3 $(mean(b_exp_3_pre))")
b_exp_4_tls = bayes_exp_4[:, 2]
b_exp_4_acc = bayes_exp_4[:, 3]
b_exp_4_vls = bayes_exp_4[:, 4]
b_exp_4_pre = bayes_exp_4[:, 5]
println("Training accuracy bayes 4 $(mean(b_exp_4_acc))")
println("Testing accuracy bayes 4 $(mean(b_exp_4_pre))")
b_exp_5_tls = bayes_exp_5[:, 2]
b_exp_5_acc = bayes_exp_5[:, 3]
b_exp_5_vls = bayes_exp_5[:, 4]
b_exp_5_pre = bayes_exp_5[:, 5]
println("Training accuracy bayes 5 $(mean(b_exp_5_acc))")
println("Testing accuracy bayes 5 $(mean(b_exp_5_pre))")
println("Amount of epochs to reach bound Bayesian 1x: $(size(bayes_exp_1)[1])")
println("Amount of epochs to reach bound Bayesian 2x: $(size(bayes_exp_2)[1])")
println("Amount of epochs to reach bound Bayesian 3x: $(size(bayes_exp_3)[1])")
println("Amount of epochs to reach bound Bayesian 4x: $(size(bayes_exp_4)[1])")
println("Amount of epochs to reach bound Bayesian 5x: $(size(bayes_exp_5)[1])")
cnn_exp_1 = load_pickle("$(folder)freq_exp_data_1.pkl");
cnn_exp_2 = load_pickle("$(folder)freq_exp_data_2.pkl");
cnn_exp_3 = load_pickle("$(folder)freq_exp_data_3.pkl");
cnn_exp_4 = load_pickle("$(folder)freq_exp_data_4.pkl");
cnn_exp_5 = load_pickle("$(folder)freq_exp_data_5.pkl");
println("Amount of epochs to reach bound LeNet 1x: $(size(cnn_exp_1)[1])")
println("Amount of epochs to reach bound LeNet 2x: $(size(cnn_exp_2)[1])")
println("Amount of epochs to reach bound LeNet 3x: $(size(cnn_exp_3)[1])")
println("Amount of epochs to reach bound LeNet 4x: $(size(cnn_exp_4)[1])")
println("Amount of epochs to reach bound LeNet 5x: $(size(cnn_exp_5)[1])")
f_exp_1_tls = cnn_exp_1[:, 2]
f_exp_1_acc = cnn_exp_1[:, 3]
f_exp_1_vls = cnn_exp_1[:, 4]
f_exp_1_pre = cnn_exp_1[:, 5]
println("Training accuracy freq 1 $(mean(f_exp_1_acc))")
println("Testing accuracy freq 1 $(mean(f_exp_1_pre))")
f_exp_2_tls = cnn_exp_2[:, 2]
f_exp_2_acc = cnn_exp_2[:, 3]
f_exp_2_vls = cnn_exp_2[:, 4]
f_exp_2_pre = cnn_exp_2[:, 5]
println("Training accuracy freq 2 $(mean(f_exp_2_acc))")
println("Testing accuracy freq 2 $(mean(f_exp_2_pre))")
f_exp_3_tls = cnn_exp_3[:, 2]
f_exp_3_acc = cnn_exp_3[:, 3]
f_exp_3_vls = cnn_exp_3[:, 4]
f_exp_3_pre = cnn_exp_3[:, 5]
println("Training accuracy freq 3 $(mean(f_exp_3_acc))")
println("Testing accuracy freq 3 $(mean(f_exp_3_pre))")
f_exp_4_tls = cnn_exp_4[:, 2]
f_exp_4_acc = cnn_exp_4[:, 3]
f_exp_4_vls = cnn_exp_4[:, 4]
f_exp_4_pre = cnn_exp_4[:, 5]
println("Training accuracy freq 4 $(mean(f_exp_4_acc))")
println("Testing accuracy freq 4 $(mean(f_exp_4_pre))")
f_exp_5_tls = cnn_exp_5[:, 2]
f_exp_5_acc = cnn_exp_5[:, 3]
f_exp_5_vls = cnn_exp_5[:, 4]
f_exp_5_pre = cnn_exp_5[:, 5]
println("Training accuracy freq 5 $(mean(f_exp_5_acc))")
println("Testing accuracy freq 5 $(mean(f_exp_5_pre))")
en_plot = plot(
[
scatter(
y = f_exp_1_acc,
name = "LeNet 1",
marker = attr(color = "rgb(211,120,000)"),
),
scatter(
y = f_exp_2_acc,
name = "LeNet 2",
marker = attr(color = "rgb(255,170,017)"),
),
scatter(
y = f_exp_3_acc,
name = "LeNet 3",
marker = attr(color = "rgb(255,187,034)"),
),
scatter(
y = f_exp_4_acc,
name = "LeNet 4",
marker = attr(color = "rgb(255,204,051)"),
),
scatter(
y = f_exp_5_acc,
name = "LeNet 5",
marker = attr(color = "rgb(255,221,068)"),
),
scatter(
y = b_exp_1_acc,
name = "BCNN 1",
marker = attr(color = "rgb(055,033,240)"),
),
scatter(
y = b_exp_2_acc,
name = "BCNN 2",
marker = attr(color = "rgb(033,081,240)"),
),
scatter(
y = b_exp_3_acc,
name = "BCNN 3",
marker = attr(color = "rgb(033,115,240)"),
),
scatter(
y = b_exp_4_acc,
name = "BCNN 4",
marker = attr(color = "rgb(151,177,255)"),
),
scatter(
y = b_exp_5_acc,
name = "BCNN 5",
marker = attr(color = "rgb(051,215,255)"),
),
],
Layout(
mode = "lines",
opacity = 0.4,
xaxis_tickangle = -45,
yaxis_title_text = "Accuracy",
xaxis_title_text = "Epoch";
yaxis_range = [0, 1],
),
)
savefig(en_plot, "cifar_ab_tacc.png")
en_plot = plot(
[
scatter(
y = f_exp_1_pre,
name = "LeNet 1",
marker = attr(color = "rgb(211,120,000)"),
),
scatter(
y = f_exp_2_pre,
name = "LeNet 2",
marker = attr(color = "rgb(255,170,017)"),
),
scatter(
y = f_exp_3_pre,
name = "LeNet 3",
marker = attr(color = "rgb(255,187,034)"),
),
scatter(
y = f_exp_4_pre,
name = "LeNet 4",
marker = attr(color = "rgb(255,204,051)"),
),
scatter(
y = f_exp_5_pre,
name = "LeNet 5",
marker = attr(color = "rgb(255,221,068)"),
),
scatter(
y = b_exp_1_pre,
name = "BCNN 1",
marker = attr(color = "rgb(055,033,240)"),
),
scatter(
y = b_exp_2_pre,
name = "BCNN 2",
marker = attr(color = "rgb(033,081,240)"),
),
scatter(
y = b_exp_3_pre,
name = "BCNN 3",
marker = attr(color = "rgb(033,115,240)"),
),
scatter(
y = b_exp_4_pre,
name = "BCNN 4",
marker = attr(color = "rgb(151,177,255)"),
),
scatter(
y = b_exp_5_pre,
name = "BCNN 5",
marker = attr(color = "rgb(051,215,255)"),
),
],
Layout(
mode = "lines",
opacity = 0.4,
xaxis_tickangle = -45,
yaxis_title_text = "Accuracy",
xaxis_title_text = "Epoch";
yaxis_range = [0, 1],
),
)
savefig(en_plot, "cifar_ab_tpre.png")

235
cifar_budget.jl Executable file
View File

@ -0,0 +1,235 @@
####################################################
#
# Energy is bounded
#
#####################################################
include("aux_func.jl")
using Statistics
using PlotlyJS
folder = "CIFAR_energy_bound/"
bayes_exp_1 = load_pickle("$(folder)bayes_exp_data_1.pkl");
bayes_exp_2 = load_pickle("$(folder)bayes_exp_data_2.pkl");
bayes_exp_3 = load_pickle("$(folder)bayes_exp_data_3.pkl");
bayes_exp_4 = load_pickle("$(folder)bayes_exp_data_4.pkl");
bayes_exp_5 = load_pickle("$(folder)bayes_exp_data_5.pkl");
b_exp_1_tls = bayes_exp_1[:, 2]
b_exp_1_acc = bayes_exp_1[:, 3]
b_exp_1_vls = bayes_exp_1[:, 4]
b_exp_1_pre = bayes_exp_1[:, 5]
println("Training accuracy bayes 1 $(mean(b_exp_1_acc))")
println("Testing accuracy bayes 1 $(mean(b_exp_1_pre))")
b_exp_2_tls = bayes_exp_2[:, 2]
b_exp_2_acc = bayes_exp_2[:, 3]
b_exp_2_vls = bayes_exp_2[:, 4]
b_exp_2_pre = bayes_exp_2[:, 5]
println("Training accuracy bayes 2 $(mean(b_exp_2_acc))")
println("Testing accuracy bayes 2 $(mean(b_exp_2_pre))")
b_exp_3_tls = bayes_exp_3[:, 2]
b_exp_3_acc = bayes_exp_3[:, 3]
b_exp_3_vls = bayes_exp_3[:, 4]
b_exp_3_pre = bayes_exp_3[:, 5]
println("Training accuracy bayes 3 $(mean(b_exp_3_acc))")
println("Testing accuracy bayes 3 $(mean(b_exp_3_pre))")
b_exp_4_tls = bayes_exp_4[:, 2]
b_exp_4_acc = bayes_exp_4[:, 3]
b_exp_4_vls = bayes_exp_4[:, 4]
b_exp_4_pre = bayes_exp_4[:, 5]
println("Training accuracy bayes 4 $(mean(b_exp_4_acc))")
println("Testing accuracy bayes 4 $(mean(b_exp_4_pre))")
b_exp_5_tls = bayes_exp_5[:, 2]
b_exp_5_acc = bayes_exp_5[:, 3]
b_exp_5_vls = bayes_exp_5[:, 4]
b_exp_5_pre = bayes_exp_5[:, 5]
println("Training accuracy bayes 5 $(mean(b_exp_5_acc))")
println("Testing accuracy bayes 5 $(mean(b_exp_5_pre))")
println("Amount of epochs to reach bound Bayesian 1x: $(size(bayes_exp_1)[1])")
println("Amount of epochs to reach bound Bayesian 2x: $(size(bayes_exp_2)[1])")
println("Amount of epochs to reach bound Bayesian 3x: $(size(bayes_exp_3)[1])")
println("Amount of epochs to reach bound Bayesian 4x: $(size(bayes_exp_4)[1])")
println("Amount of epochs to reach bound Bayesian 5x: $(size(bayes_exp_5)[1])")
cnn_exp_1 = load_pickle("$(folder)freq_exp_data_1.pkl");
cnn_exp_2 = load_pickle("$(folder)freq_exp_data_2.pkl");
cnn_exp_3 = load_pickle("$(folder)freq_exp_data_3.pkl");
cnn_exp_4 = load_pickle("$(folder)freq_exp_data_4.pkl");
cnn_exp_5 = load_pickle("$(folder)freq_exp_data_5.pkl");
f_exp_1_tls = cnn_exp_1[:, 2]
f_exp_1_acc = cnn_exp_1[:, 3]
f_exp_1_vls = cnn_exp_1[:, 4]
f_exp_1_pre = cnn_exp_1[:, 5]
println("Training accuracy freq 1 $(mean(f_exp_1_acc))")
println("Testing accuracy freq 1 $(mean(f_exp_1_pre))")
f_exp_2_tls = cnn_exp_2[:, 2]
f_exp_2_acc = cnn_exp_2[:, 3]
f_exp_2_vls = cnn_exp_2[:, 4]
f_exp_2_pre = cnn_exp_2[:, 5]
println("Training accuracy freq 2 $(mean(f_exp_2_acc))")
println("Testing accuracy freq 2 $(mean(f_exp_2_pre))")
f_exp_3_tls = cnn_exp_3[:, 2]
f_exp_3_acc = cnn_exp_3[:, 3]
f_exp_3_vls = cnn_exp_3[:, 4]
f_exp_3_pre = cnn_exp_3[:, 5]
println("Training accuracy freq 3 $(mean(f_exp_3_acc))")
println("Testing accuracy freq 3 $(mean(f_exp_3_pre))")
f_exp_4_tls = cnn_exp_4[:, 2]
f_exp_4_acc = cnn_exp_4[:, 3]
f_exp_4_vls = cnn_exp_4[:, 4]
f_exp_4_pre = cnn_exp_4[:, 5]
println("Training accuracy freq 4 $(mean(f_exp_4_acc))")
println("Testing accuracy freq 4 $(mean(f_exp_4_pre))")
f_exp_5_tls = cnn_exp_5[:, 2]
f_exp_5_acc = cnn_exp_5[:, 3]
f_exp_5_vls = cnn_exp_5[:, 4]
f_exp_5_pre = cnn_exp_5[:, 5]
println("Training accuracy freq 5 $(mean(f_exp_5_acc))")
println("Testing accuracy freq 5 $(mean(f_exp_5_pre))")
println("Amount of epochs to reach bound LeNet 1x: $(size(cnn_exp_1)[1])")
println("Amount of epochs to reach bound LeNet 2x: $(size(cnn_exp_2)[1])")
println("Amount of epochs to reach bound LeNet 3x: $(size(cnn_exp_3)[1])")
println("Amount of epochs to reach bound LeNet 4x: $(size(cnn_exp_4)[1])")
println("Amount of epochs to reach bound LeNet 5x: $(size(cnn_exp_5)[1])")
en_plot = plot(
[
scatter(
y = f_exp_1_acc,
name = "LeNet 1",
marker = attr(color = "rgb(211,120,000)"),
),
scatter(
y = f_exp_2_acc,
name = "LeNet 2",
marker = attr(color = "rgb(255,170,017)"),
),
scatter(
y = f_exp_3_acc,
name = "LeNet 3",
marker = attr(color = "rgb(255,187,034)"),
),
scatter(
y = f_exp_4_acc,
name = "LeNet 4",
marker = attr(color = "rgb(255,204,051)"),
),
scatter(
y = f_exp_5_acc,
name = "LeNet 5",
marker = attr(color = "rgb(255,221,068)"),
),
scatter(
y = b_exp_1_acc,
name = "BCNN 1",
marker = attr(color = "rgb(055,033,240)"),
),
scatter(
y = b_exp_2_acc,
name = "BCNN 2",
marker = attr(color = "rgb(033,081,240)"),
),
scatter(
y = b_exp_3_acc,
name = "BCNN 3",
marker = attr(color = "rgb(033,115,240)"),
),
scatter(
y = b_exp_4_acc,
name = "BCNN 4",
marker = attr(color = "rgb(151,177,255)"),
),
scatter(
y = b_exp_5_acc,
name = "BCNN 5",
marker = attr(color = "rgb(051,215,255)"),
),
],
Layout(
mode = "lines",
opacity = 0.4,
xaxis_tickangle = -45,
yaxis_title_text = "Accuracy",
xaxis_title_text = "Epoch";
yaxis_range = [0, 1],
),
)
savefig(en_plot, "cifar_eb_tacc.png")
en_plot = plot(
[
scatter(
y = f_exp_1_pre,
name = "LeNet 1",
marker = attr(color = "rgb(211,120,000)"),
),
scatter(
y = f_exp_2_pre,
name = "LeNet 2",
marker = attr(color = "rgb(255,170,017)"),
),
scatter(
y = f_exp_3_pre,
name = "LeNet 3",
marker = attr(color = "rgb(255,187,034)"),
),
scatter(
y = f_exp_4_pre,
name = "LeNet 4",
marker = attr(color = "rgb(255,204,051)"),
),
scatter(
y = f_exp_5_pre,
name = "LeNet 5",
marker = attr(color = "rgb(255,221,068)"),
),
scatter(
y = b_exp_1_pre,
name = "BCNN 1",
marker = attr(color = "rgb(055,033,240)"),
),
scatter(
y = b_exp_2_pre,
name = "BCNN 2",
marker = attr(color = "rgb(033,081,240)"),
),
scatter(
y = b_exp_3_pre,
name = "BCNN 3",
marker = attr(color = "rgb(033,115,240)"),
),
scatter(
y = b_exp_4_pre,
name = "BCNN 4",
marker = attr(color = "rgb(151,177,255)"),
),
scatter(
y = b_exp_5_pre,
name = "BCNN 5",
marker = attr(color = "rgb(051,215,255)"),
),
],
Layout(
mode = "lines",
opacity = 0.4,
xaxis_tickangle = -45,
yaxis_title_text = "Accuracy",
xaxis_title_text = "Epoch";
yaxis_range = [0, 1],
),
)
savefig(en_plot, "cifar_eb_tpre.png")

230
cifar_early_exp.jl Executable file
View File

@ -0,0 +1,230 @@
include("aux_func.jl")
using Statistics
using PlotlyJS
using PlotlyJS: savefig
folder = "CIFAR_early_stop/"
bayes_exp_1 = load_pickle("$(folder)bayes_exp_data_1.pkl")
bayes_exp_2 = load_pickle("$(folder)bayes_exp_data_2.pkl")
bayes_exp_3 = load_pickle("$(folder)bayes_exp_data_3.pkl")
bayes_exp_4 = load_pickle("$(folder)bayes_exp_data_4.pkl")
bayes_exp_5 = load_pickle("$(folder)bayes_exp_data_5.pkl")
b_exp_1_tls = bayes_exp_1[:, 2]
b_exp_1_acc = bayes_exp_1[:, 3]
b_exp_1_vls = bayes_exp_1[:, 4]
b_exp_1_pre = bayes_exp_1[:, 5]
println("Training accuracy bayes 1 $(mean(b_exp_1_acc))")
println("Testing accuracy bayes 1 $(mean(b_exp_1_pre))")
b_exp_2_tls = bayes_exp_2[:, 2]
b_exp_2_acc = bayes_exp_2[:, 3]
b_exp_2_vls = bayes_exp_2[:, 4]
b_exp_2_pre = bayes_exp_2[:, 5]
println("Training accuracy bayes 2 $(mean(b_exp_2_acc))")
println("Testing accuracy bayes 2 $(mean(b_exp_2_pre))")
b_exp_3_tls = bayes_exp_3[:, 2]
b_exp_3_acc = bayes_exp_3[:, 3]
b_exp_3_vls = bayes_exp_3[:, 4]
b_exp_3_pre = bayes_exp_3[:, 5]
println("Training accuracy bayes 3 $(mean(b_exp_3_acc))")
println("Testing accuracy bayes 3 $(mean(b_exp_3_pre))")
b_exp_4_tls = bayes_exp_4[:, 2]
b_exp_4_acc = bayes_exp_4[:, 3]
b_exp_4_vls = bayes_exp_4[:, 4]
b_exp_4_pre = bayes_exp_4[:, 5]
println("Training accuracy bayes 4 $(mean(b_exp_4_acc))")
println("Testing accuracy bayes 4 $(mean(b_exp_4_pre))")
b_exp_5_tls = bayes_exp_5[:, 2]
b_exp_5_acc = bayes_exp_5[:, 3]
b_exp_5_vls = bayes_exp_5[:, 4]
b_exp_5_pre = bayes_exp_5[:, 5]
println("Training accuracy bayes 5 $(mean(b_exp_5_acc))")
println("Testing accuracy bayes 5 $(mean(b_exp_5_pre))")
println("Amount of epochs to reach bound Bayesian 1x: $(size(bayes_exp_1)[1])")
println("Amount of epochs to reach bound Bayesian 2x: $(size(bayes_exp_2)[1])")
println("Amount of epochs to reach bound Bayesian 3x: $(size(bayes_exp_3)[1])")
println("Amount of epochs to reach bound Bayesian 4x: $(size(bayes_exp_4)[1])")
println("Amount of epochs to reach bound Bayesian 5x: $(size(bayes_exp_5)[1])")
cnn_exp_1 = load_pickle("$(folder)freq_exp_data_1.pkl")
cnn_exp_2 = load_pickle("$(folder)freq_exp_data_2.pkl")
cnn_exp_3 = load_pickle("$(folder)freq_exp_data_3.pkl")
cnn_exp_4 = load_pickle("$(folder)freq_exp_data_4.pkl")
cnn_exp_5 = load_pickle("$(folder)freq_exp_data_5.pkl")
f_exp_1_tls = cnn_exp_1[:, 2]
f_exp_1_acc = cnn_exp_1[:, 3]
f_exp_1_vls = cnn_exp_1[:, 4]
f_exp_1_pre = cnn_exp_1[:, 5]
println("Training accuracy freq 1 $(mean(f_exp_1_acc))")
println("Testing accuracy freq 1 $(mean(f_exp_1_pre))")
f_exp_2_tls = cnn_exp_2[:, 2]
f_exp_2_acc = cnn_exp_2[:, 3]
f_exp_2_vls = cnn_exp_2[:, 4]
f_exp_2_pre = cnn_exp_2[:, 5]
println("Training accuracy freq 2 $(mean(f_exp_2_acc))")
println("Testing accuracy freq 2 $(mean(f_exp_2_pre))")
f_exp_3_tls = cnn_exp_3[:, 2]
f_exp_3_acc = cnn_exp_3[:, 3]
f_exp_3_vls = cnn_exp_3[:, 4]
f_exp_3_pre = cnn_exp_3[:, 5]
println("Training accuracy freq 3 $(mean(f_exp_3_acc))")
println("Testing accuracy freq 3 $(mean(f_exp_3_pre))")
f_exp_4_tls = cnn_exp_4[:, 2]
f_exp_4_acc = cnn_exp_4[:, 3]
f_exp_4_vls = cnn_exp_4[:, 4]
f_exp_4_pre = cnn_exp_4[:, 5]
println("Training accuracy freq 4 $(mean(f_exp_4_acc))")
println("Testing accuracy freq 4 $(mean(f_exp_4_pre))")
f_exp_5_tls = cnn_exp_5[:, 2]
f_exp_5_acc = cnn_exp_5[:, 3]
f_exp_5_vls = cnn_exp_5[:, 4]
f_exp_5_pre = cnn_exp_5[:, 5]
println("Training accuracy freq 5 $(mean(f_exp_5_acc))")
println("Testing accuracy freq 5 $(mean(f_exp_5_pre))")
println("Amount of epochs to reach bound LeNet 1x: $(size(cnn_exp_1)[1])")
println("Amount of epochs to reach bound LeNet 2x: $(size(cnn_exp_2)[1])")
println("Amount of epochs to reach bound LeNet 3x: $(size(cnn_exp_3)[1])")
println("Amount of epochs to reach bound LeNet 4x: $(size(cnn_exp_4)[1])")
println("Amount of epochs to reach bound LeNet 5x: $(size(cnn_exp_5)[1])")
en_plot = plot(
[
scatter(
y = f_exp_1_acc,
name = "LeNet 1",
marker = attr(color = "rgb(211,120,000)"),
),
scatter(
y = f_exp_2_acc,
name = "LeNet 2",
marker = attr(color = "rgb(255,170,017)"),
),
scatter(
y = f_exp_3_acc,
name = "LeNet 3",
marker = attr(color = "rgb(255,187,034)"),
),
scatter(
y = f_exp_4_acc,
name = "LeNet 4",
marker = attr(color = "rgb(255,204,051)"),
),
scatter(
y = f_exp_5_acc,
name = "LeNet 5",
marker = attr(color = "rgb(255,221,068)"),
),
scatter(
y = b_exp_1_acc,
name = "BCNN 1",
marker = attr(color = "rgb(055,033,240)"),
),
scatter(
y = b_exp_2_acc,
name = "BCNN 2",
marker = attr(color = "rgb(033,081,240)"),
),
scatter(
y = b_exp_3_acc,
name = "BCNN 3",
marker = attr(color = "rgb(033,115,240)"),
),
scatter(
y = b_exp_4_acc,
name = "BCNN 4",
marker = attr(color = "rgb(151,177,255)"),
),
scatter(
y = b_exp_5_acc,
name = "BCNN 5",
marker = attr(color = "rgb(051,215,255)"),
),
],
Layout(
mode = "lines",
opacity = 0.4,
xaxis_tickangle = -45,
yaxis_title_text = "Accuracy",
xaxis_title_text = "Epoch";
yaxis_range = [0, 1],
),
)
savefig(en_plot, "cifar_es_tacc.png")
en_plot = plot(
[
scatter(
y = f_exp_1_pre,
name = "LeNet 1",
marker = attr(color = "rgb(211,120,000)"),
),
scatter(
y = f_exp_2_pre,
name = "LeNet 2",
marker = attr(color = "rgb(255,170,017)"),
),
scatter(
y = f_exp_3_pre,
name = "LeNet 3",
marker = attr(color = "rgb(255,187,034)"),
),
scatter(
y = f_exp_4_pre,
name = "LeNet 4",
marker = attr(color = "rgb(255,204,051)"),
),
scatter(
y = f_exp_5_pre,
name = "LeNet 5",
marker = attr(color = "rgb(255,221,068)"),
),
scatter(
y = b_exp_1_pre,
name = "BCNN 1",
marker = attr(color = "rgb(055,033,240)"),
),
scatter(
y = b_exp_2_pre,
name = "BCNN 2",
marker = attr(color = "rgb(033,081,240)"),
),
scatter(
y = b_exp_3_pre,
name = "BCNN 3",
marker = attr(color = "rgb(033,115,240)"),
),
scatter(
y = b_exp_4_pre,
name = "BCNN 4",
marker = attr(color = "rgb(151,177,255)"),
),
scatter(
y = b_exp_5_pre,
name = "BCNN 5",
marker = attr(color = "rgb(051,215,255)"),
),
],
Layout(
mode = "lines",
opacity = 0.4,
xaxis_tickangle = -45,
yaxis_title_text = "Accuracy",
xaxis_title_text = "Epoch";
yaxis_range = [0, 1],
),
)
savefig(en_plot, "cifar_es_tpre.png")

310
cpu_results.jl Executable file
View File

@ -0,0 +1,310 @@
include("aux_func.jl")
using Statistics
using Plots
cif = "cifar";
mni = "mnist";
bay = "bayes";
wat = "cpu_watts";
frq = "freq";
ram = "ram_use";
_100 = "100";
acc = "acc";
es = "es";
wbud = "wbud";
bay_cif_acc_ene = readcpudata(cif, bay, wat, acc);
bay_cif_acc_ram = readcpudata(cif, bay, ram, acc);
bay_cif_es_ene = readcpudata(cif, bay, wat, es);
bay_cif_es_ram = readcpudata(cif, bay, ram, es);
bay_cif_wbud_ene = readcpudata(cif, bay, wat, wbud);
bay_cif_wbud_ram = readcpudata(cif, bay, ram, wbud);
bay_cif_100_ene = readcpudata(cif, bay, wat, _100);
bay_cif_100_ram = readcpudata(cif, bay, ram, _100);
bay_mni_acc_ene = readcpudata(mni, bay, wat, acc);
bay_mni_acc_ram = readcpudata(mni, bay, ram, acc);
bay_mni_es_ene = readcpudata(mni, bay, wat, es);
bay_mni_es_ram = readcpudata(mni, bay, ram, es);
bay_mni_wbud_ene = readcpudata(mni, bay, wat, wbud);
bay_mni_wbud_ram = readcpudata(mni, bay, ram, wbud);
bay_mni_100_ene = readcpudata(mni, bay, wat, _100);
bay_mni_100_ram = readcpudata(mni, bay, ram, _100);
frq_cif_acc_ene = readcpudata(cif, frq, wat, acc);
frq_cif_acc_ram = readcpudata(cif, frq, ram, acc);
frq_cif_es_ene = readcpudata(cif, frq, wat, es);
frq_cif_es_ram = readcpudata(cif, frq, ram, es);
frq_cif_wbud_ene = readcpudata(cif, frq, wat, wbud);
frq_cif_wbud_ram = readcpudata(cif, frq, ram, wbud);
frq_cif_100_ene = readcpudata(cif, frq, wat, _100);
frq_cif_100_ram = readcpudata(cif, frq, ram, _100);
frq_mni_acc_ene = readcpudata(mni, frq, wat, acc);
frq_mni_acc_ram = readcpudata(mni, frq, ram, acc);
frq_mni_es_ene = readcpudata(mni, frq, wat, es);
frq_mni_es_ram = readcpudata(mni, frq, ram, es);
frq_mni_wbud_ene = readcpudata(mni, frq, wat, wbud);
frq_mni_wbud_ram = readcpudata(mni, frq, ram, wbud);
frq_mni_100_ene = readcpudata(mni, frq, wat, _100);
frq_mni_100_ram = readcpudata(mni, frq, ram, _100);
temp = Vector()
for vec in bay_cif_acc_ene
t = getcpuwatt(vec)
push!(temp, t)
end
bay_cif_acc_ene = temp;
temp = Vector()
for vec in bay_cif_acc_ram
t = getramuse(vec)
push!(temp, t)
end
bay_cif_acc_ram = temp;
temp = Vector()
for vec in bay_cif_es_ene
t = getcpuwatt(vec)
push!(temp, t)
end
bay_cif_es_ene = temp;
temp = Vector()
for vec in bay_cif_es_ram
t = getramuse(vec)
push!(temp, t)
end
bay_cif_es_ram = temp;
temp = Vector()
for vec in bay_cif_wbud_ene
t = getcpuwatt(vec)
push!(temp, t)
end
bay_cif_wbud_ene = temp;
temp = Vector()
for vec in bay_cif_wbud_ram
t = getramuse(vec)
push!(temp, t)
end
bay_cif_wbud_ram = temp;
temp = Vector()
for vec in bay_cif_100_ene
t = getcpuwatt(vec)
push!(temp, t)
end
bay_cif_100_ene = temp;
temp = Vector()
for vec in bay_cif_100_ram
t = getramuse(vec)
push!(temp, t)
end
bay_cif_100_ram = temp;
temp = Vector()
for vec in bay_mni_acc_ene
t = getcpuwatt(vec)
push!(temp, t)
end
bay_mni_acc_ene = temp;
temp = Vector()
for vec in bay_mni_acc_ram
t = getramuse(vec)
push!(temp, t)
end
bay_mni_acc_ram = temp;
temp = Vector()
for vec in bay_mni_es_ene
t = getcpuwatt(vec)
push!(temp, t)
end
bay_mni_es_ene = temp;
temp = Vector()
for vec in bay_mni_es_ram
t = getramuse(vec)
push!(temp, t)
end
bay_mni_es_ram = temp;
temp = Vector()
for vec in bay_mni_wbud_ene
t = getcpuwatt(vec)
push!(temp, t)
end
bay_mni_wbud_ene = temp;
temp = Vector()
for vec in bay_mni_wbud_ram
t = getramuse(vec)
push!(temp, t)
end
bay_mni_wbud_ram = temp;
temp = Vector()
for vec in bay_mni_100_ene
t = getcpuwatt(vec)
push!(temp, t)
end
bay_mni_100_ene = temp;
temp = Vector()
for vec in bay_mni_100_ram
t = getramuse(vec)
push!(temp, t)
end
bay_mni_100_ram = temp;
temp = Vector()
for vec in frq_cif_acc_ene
t = getcpuwatt(vec)
push!(temp, t)
end
frq_cif_acc_ene = temp;
temp = Vector()
for vec in frq_cif_acc_ram
t = getramuse(vec)
push!(temp, t)
end
frq_cif_acc_ram = temp;
temp = Vector()
for vec in frq_cif_es_ene
t = getcpuwatt(vec)
push!(temp, t)
end
frq_cif_es_ene = temp;
temp = Vector()
for vec in frq_cif_es_ram
t = getramuse(vec)
push!(temp, t)
end
frq_cif_es_ram = temp;
temp = Vector()
for vec in frq_cif_wbud_ene
t = getcpuwatt(vec)
push!(temp, t)
end
frq_cif_wbud_ene = temp;
temp = Vector()
for vec in frq_cif_wbud_ram
t = getramuse(vec)
push!(temp, t)
end
frq_cif_wbud_ram = temp;
temp = Vector()
for vec in frq_cif_100_ene
t = getcpuwatt(vec)
push!(temp, t)
end
frq_cif_100_ene = temp;
temp = Vector()
for vec in frq_cif_100_ram
t = getramuse(vec)
push!(temp, t)
end
frq_cif_100_ram = temp;
temp = Vector()
for vec in frq_mni_acc_ene
t = getcpuwatt(vec)
push!(temp, t)
end
frq_mni_acc_ene = temp;
temp = Vector()
for vec in frq_mni_acc_ram
t = getramuse(vec)
push!(temp, t)
end
frq_mni_acc_ram = temp;
temp = Vector()
for vec in frq_mni_es_ene
t = getcpuwatt(vec)
push!(temp, t)
end
frq_mni_es_ene = temp;
temp = Vector()
for vec in frq_mni_es_ram
t = getramuse(vec)
push!(temp, t)
end
frq_mni_es_ram = temp;
temp = Vector()
for vec in frq_mni_wbud_ene
t = getcpuwatt(vec)
push!(temp, t)
end
frq_mni_wbud_ene = temp;
temp = Vector()
for vec in frq_mni_wbud_ram
t = getramuse(vec)
push!(temp, t)
end
frq_mni_wbud_ram = temp;
temp = Vector()
for vec in frq_mni_100_ene
t = getcpuwatt(vec)
push!(temp, t)
end
frq_mni_100_ene = temp;
temp = Vector()
for vec in frq_mni_100_ram
t = getramuse(vec)
push!(temp, t)
end
frq_mni_100_ram = temp;

23
datafft.jl Executable file
View File

@ -0,0 +1,23 @@
include("aux_func.jl")
using Plots
using FFTW
using DSP
folder = "exp_100_epochs/"
bay_watt_1 = load_pickle("$(folder)bayes_wattdata_1.pkl")
bay_watt_1 = removewatt(bay_watt_1[:, 1])
N = length(bay_watt_1) - 1
Ts = 1 / (1.1 * N)
t0 = 0
tmax = t0 + N * Ts
t = t0:Ts:tmax
bay_watt_1_fft = fft(bay_watt_1) |> fftshift
bay_watt_1_fft_f = fftfreq(length(t), 1.0 / Ts) |> fftshift
plot(bay_watt_1)
plot(bay_watt_1_fft_f, abs.(bay_watt_1_fft), title = "Spectrum")
length(t)

233
early_exp.jl Executable file
View File

@ -0,0 +1,233 @@
include("aux_func.jl")
using Statistics
using PlotlyJS
using PlotlyJS: savefig
folder = "early_stop_res/"
bayes_exp_1 = load_pickle("$(folder)bayes_exp_data_1.pkl")
bayes_exp_2 = load_pickle("$(folder)bayes_exp_data_2.pkl")
bayes_exp_3 = load_pickle("$(folder)bayes_exp_data_3.pkl")
bayes_exp_4 = load_pickle("$(folder)bayes_exp_data_4.pkl")
bayes_exp_5 = load_pickle("$(folder)bayes_exp_data_5.pkl")
b_exp_1_tls = bayes_exp_1[:, 2]
b_exp_1_acc = bayes_exp_1[:, 3]
b_exp_1_vls = bayes_exp_1[:, 4]
b_exp_1_pre = bayes_exp_1[:, 5]
println("Training accuracy bayes 1 $(mean(b_exp_1_acc))")
println("Testing accuracy bayes 1 $(mean(b_exp_1_pre))")
b_exp_2_tls = bayes_exp_2[:, 2]
b_exp_2_acc = bayes_exp_2[:, 3]
b_exp_2_vls = bayes_exp_2[:, 4]
b_exp_2_pre = bayes_exp_2[:, 5]
println("Training accuracy bayes 2 $(mean(b_exp_2_acc))")
println("Testing accuracy bayes 2 $(mean(b_exp_2_pre))")
b_exp_3_tls = bayes_exp_3[:, 2]
b_exp_3_acc = bayes_exp_3[:, 3]
b_exp_3_vls = bayes_exp_3[:, 4]
b_exp_3_pre = bayes_exp_3[:, 5]
println("Training accuracy bayes 3 $(mean(b_exp_3_acc))")
println("Testing accuracy bayes 3 $(mean(b_exp_3_pre))")
b_exp_4_tls = bayes_exp_4[:, 2]
b_exp_4_acc = bayes_exp_4[:, 3]
b_exp_4_vls = bayes_exp_4[:, 4]
b_exp_4_pre = bayes_exp_4[:, 5]
println("Training accuracy bayes 4 $(mean(b_exp_4_acc))")
println("Testing accuracy bayes 4 $(mean(b_exp_4_pre))")
b_exp_5_tls = bayes_exp_5[:, 2]
b_exp_5_acc = bayes_exp_5[:, 3]
b_exp_5_vls = bayes_exp_5[:, 4]
b_exp_5_pre = bayes_exp_5[:, 5]
println("Training accuracy bayes 5 $(mean(b_exp_5_acc))")
println("Testing accuracy bayes 5 $(mean(b_exp_5_pre))")
println("Amount of epochs to reach bound Bayesian 1x: $(size(bayes_exp_1)[1])")
println("Amount of epochs to reach bound Bayesian 2x: $(size(bayes_exp_2)[1])")
println("Amount of epochs to reach bound Bayesian 3x: $(size(bayes_exp_3)[1])")
println("Amount of epochs to reach bound Bayesian 4x: $(size(bayes_exp_4)[1])")
println("Amount of epochs to reach bound Bayesian 5x: $(size(bayes_exp_5)[1])")
cnn_exp_1 = load_pickle("$(folder)freq_exp_data_1.pkl")
cnn_exp_2 = load_pickle("$(folder)freq_exp_data_2.pkl")
cnn_exp_3 = load_pickle("$(folder)freq_exp_data_3.pkl")
cnn_exp_4 = load_pickle("$(folder)freq_exp_data_4.pkl")
cnn_exp_5 = load_pickle("$(folder)freq_exp_data_5.pkl")
f_exp_1_tls = cnn_exp_1[:, 2]
f_exp_1_acc = cnn_exp_1[:, 3]
f_exp_1_vls = cnn_exp_1[:, 4]
f_exp_1_pre = cnn_exp_1[:, 5]
println("Training accuracy freq 1 $(mean(f_exp_1_acc))")
println("Testing accuracy freq 1 $(mean(f_exp_1_pre))")
f_exp_2_tls = cnn_exp_2[:, 2]
f_exp_2_acc = cnn_exp_2[:, 3]
f_exp_2_vls = cnn_exp_2[:, 4]
f_exp_2_pre = cnn_exp_2[:, 5]
println("Training accuracy freq 2 $(mean(f_exp_2_acc))")
println("Testing accuracy freq 2 $(mean(f_exp_2_pre))")
f_exp_3_tls = cnn_exp_3[:, 2]
f_exp_3_acc = cnn_exp_3[:, 3]
f_exp_3_vls = cnn_exp_3[:, 4]
f_exp_3_pre = cnn_exp_3[:, 5]
println("Training accuracy freq 3 $(mean(f_exp_3_acc))")
println("Testing accuracy freq 3 $(mean(f_exp_3_pre))")
f_exp_4_tls = cnn_exp_4[:, 2]
f_exp_4_acc = cnn_exp_4[:, 3]
f_exp_4_vls = cnn_exp_4[:, 4]
f_exp_4_pre = cnn_exp_4[:, 5]
println("Training accuracy freq 4 $(mean(f_exp_4_acc))")
println("Testing accuracy freq 4 $(mean(f_exp_4_pre))")
f_exp_5_tls = cnn_exp_5[:, 2]
f_exp_5_acc = cnn_exp_5[:, 3]
f_exp_5_vls = cnn_exp_5[:, 4]
f_exp_5_pre = cnn_exp_5[:, 5]
println("Training accuracy freq 5 $(mean(f_exp_5_acc))")
println("Testing accuracy freq 5 $(mean(f_exp_5_pre))")
println("Amount of epochs to reach bound LeNet 1x: $(size(cnn_exp_1)[1])")
println("Amount of epochs to reach bound LeNet 2x: $(size(cnn_exp_2)[1])")
println("Amount of epochs to reach bound LeNet 3x: $(size(cnn_exp_3)[1])")
println("Amount of epochs to reach bound LeNet 4x: $(size(cnn_exp_4)[1])")
println("Amount of epochs to reach bound LeNet 5x: $(size(cnn_exp_5)[1])")
en_plot = plot(
[
scatter(
y = f_exp_1_acc,
name = "LeNet 1",
marker = attr(color = "rgb(211,120,000)"),
),
scatter(
y = f_exp_2_acc,
name = "LeNet 2",
marker = attr(color = "rgb(255,170,017)"),
),
scatter(
y = f_exp_3_acc,
name = "LeNet 3",
marker = attr(color = "rgb(255,187,034)"),
),
scatter(
y = f_exp_4_acc,
name = "LeNet 4",
marker = attr(color = "rgb(255,204,051)"),
),
scatter(
y = f_exp_5_acc,
name = "LeNet 5",
marker = attr(color = "rgb(255,221,068)"),
),
scatter(
y = b_exp_1_acc,
name = "BCNN 1",
marker = attr(color = "rgb(055,033,240)"),
),
scatter(
y = b_exp_2_acc,
name = "BCNN 2",
marker = attr(color = "rgb(033,081,240)"),
),
scatter(
y = b_exp_3_acc,
name = "BCNN 3",
marker = attr(color = "rgb(033,115,240)"),
),
scatter(
y = b_exp_4_acc,
name = "BCNN 4",
marker = attr(color = "rgb(151,177,255)"),
),
scatter(
y = b_exp_5_acc,
name = "BCNN 5",
marker = attr(color = "rgb(051,215,255)"),
),
],
Layout(
mode = "lines",
opacity = 0.4,
xaxis_tickangle = -45,
yaxis_title_text = "Accuracy",
xaxis_title_text = "Epoch";
yaxis_range = [0, 1],
),
)
savefig(en_plot, "mnist_es_tacc.png")
en_plot = plot(
[
scatter(
y = f_exp_1_pre,
name = "LeNet 1",
marker = attr(color = "rgb(211,120,000)"),
),
scatter(
y = f_exp_2_pre,
name = "LeNet 2",
marker = attr(color = "rgb(255,170,017)"),
),
scatter(
y = f_exp_3_pre,
name = "LeNet 3",
marker = attr(color = "rgb(255,187,034)"),
),
scatter(
y = f_exp_4_pre,
name = "LeNet 4",
marker = attr(color = "rgb(255,204,051)"),
),
scatter(
y = f_exp_5_pre,
name = "LeNet 5",
marker = attr(color = "rgb(255,221,068)"),
),
scatter(
y = b_exp_1_pre,
name = "BCNN 1",
marker = attr(color = "rgb(055,033,240)"),
),
scatter(
y = b_exp_2_pre,
name = "BCNN 2",
marker = attr(color = "rgb(033,081,240)"),
),
scatter(
y = b_exp_3_pre,
name = "BCNN 3",
marker = attr(color = "rgb(033,115,240)"),
),
scatter(
y = b_exp_4_pre,
name = "BCNN 4",
marker = attr(color = "rgb(151,177,255)"),
),
scatter(
y = b_exp_5_pre,
name = "BCNN 5",
marker = attr(color = "rgb(051,215,255)"),
),
],
Layout(
mode = "lines",
opacity = 0.4,
xaxis_tickangle = -45,
yaxis_title_text = "Accuracy",
xaxis_title_text = "Epoch";
yaxis_range = [0, 1],
),
)
savefig(en_plot, "mnist_es_tpre.png")

297
efficiency_graphs.jl Executable file
View File

@ -0,0 +1,297 @@
using PlotlyJS
using PlotlyJS: savefig
using Statistics: mean, std
include("aux_func.jl")
data = load_pickle("efficiency_per_size.pkl");
#all_data_ene["mni"][s]["100"]["bay"]
data_type = ["mni", "cif"]
model_type = ["bay", "frq"]
experiment_type = ["100", "est", "acc", "wat"]
# MNIST 100 efficency
bayes_100_mnist_eff = []
for s = 1:5
push!(
bayes_100_mnist_eff,
data[model_type[1]][data_type[1]][experiment_type[1]][string(s)],
)
end
freqs_100_mnist_eff = []
for s = 1:5
push!(
freqs_100_mnist_eff,
data[model_type[2]][data_type[1]][experiment_type[1]][string(s)],
)
end
# MNIST est efficency
bayes_est_mnist_eff = []
for s = 1:5
push!(
bayes_est_mnist_eff,
data[model_type[1]][data_type[1]][experiment_type[2]][string(s)],
)
end
freqs_est_mnist_eff = []
for s = 1:5
push!(
freqs_est_mnist_eff,
data[model_type[2]][data_type[1]][experiment_type[2]][string(s)],
)
end
# MNIST wat efficency
bayes_wat_mnist_eff = []
for s = 1:5
push!(
bayes_wat_mnist_eff,
data[model_type[1]][data_type[1]][experiment_type[4]][string(s)],
)
end
freqs_wat_mnist_eff = []
for s = 1:5
push!(
freqs_wat_mnist_eff,
data[model_type[2]][data_type[1]][experiment_type[4]][string(s)],
)
end
# MNIST acc efficency
bayes_acc_mnist_eff = []
for s = 1:5
push!(
bayes_acc_mnist_eff,
data[model_type[1]][data_type[1]][experiment_type[3]][string(s)],
)
end
freqs_acc_mnist_eff = []
for s = 1:5
push!(
freqs_acc_mnist_eff,
data[model_type[2]][data_type[1]][experiment_type[3]][string(s)],
)
end
# CIFAR 100 efficency
bayes_100_cifar_eff = []
for s = 1:5
push!(
bayes_100_cifar_eff,
data[model_type[1]][data_type[2]][experiment_type[1]][string(s)],
)
end
for i = 1:5
t_std = std(bayes_100_cifar_eff)
if (bayes_100_cifar_eff[i] > 2 * t_std) || (bayes_100_cifar_eff[i] < 2 * t_std)
bayes_100_cifar_eff[i] = mean(bayes_100_cifar_eff)
end
end
freqs_100_cifar_eff = []
for s = 1:5
push!(
freqs_100_cifar_eff,
data[model_type[2]][data_type[2]][experiment_type[1]][string(s)],
)
end
for i = 1:5
t_std = std(freqs_100_cifar_eff)
if (freqs_100_cifar_eff[i] > 2 * t_std) || (freqs_100_cifar_eff[i] < 2 * t_std)
freqs_100_cifar_eff[i] = mean(freqs_100_cifar_eff)
end
end
# CIFAR est efficency
bayes_est_cifar_eff = []
for s = 1:5
push!(
bayes_est_cifar_eff,
data[model_type[1]][data_type[2]][experiment_type[2]][string(s)],
)
end
freqs_est_cifar_eff = []
for s = 1:5
push!(
freqs_est_cifar_eff,
data[model_type[2]][data_type[2]][experiment_type[2]][string(s)],
)
end
# CIFAR wat efficency
bayes_wat_cifar_eff = []
for s = 1:5
push!(
bayes_wat_cifar_eff,
data[model_type[1]][data_type[2]][experiment_type[4]][string(s)],
)
end
freqs_wat_cifar_eff = []
for s = 1:5
push!(
freqs_wat_cifar_eff,
data[model_type[2]][data_type[2]][experiment_type[4]][string(s)],
)
end
# CIFAR acc efficency
bayes_acc_cifar_eff = []
for s = 1:5
push!(
bayes_acc_cifar_eff,
data[model_type[1]][data_type[2]][experiment_type[3]][string(s)],
)
end
freqs_acc_cifar_eff = []
for s = 1:5
push!(
freqs_acc_cifar_eff,
data[model_type[2]][data_type[2]][experiment_type[3]][string(s)],
)
end
avg_rate_bcnn = (-8.266684252643054e-5 * 1000)
avg_rate_fcnn = (0.00022035677966088333 * 1000)
en_plot = plot(
[
scatter(
x = ["1", "2", "3", "4", "5"],
y = (-avg_rate_fcnn .* freqs_100_mnist_eff),
name = "LeNet 100",
marker = attr(symbol = 4, color = "rgb(211,120,000)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = (-avg_rate_fcnn .* freqs_est_mnist_eff),
name = "LeNet est",
marker = attr(symbol = 17, color = "rgb(255,170,017)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = (-avg_rate_fcnn .* freqs_wat_mnist_eff),
name = "LeNet wat",
marker = attr(symbol = 2, color = "rgb(255,187,034)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = (-avg_rate_fcnn .* freqs_acc_mnist_eff),
name = "LeNet acc",
marker = attr(symbol = 0, color = "rgb(255,204,051)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = (-avg_rate_bcnn .* bayes_100_mnist_eff),
name = "BCNN 100",
marker = attr(symbol = 4, color = "rgb(055,033,240)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = (-avg_rate_bcnn .* bayes_est_mnist_eff),
name = "BCNN est",
marker = attr(symbol = 17, color = "rgb(033,081,240)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = (-avg_rate_bcnn .* bayes_wat_mnist_eff),
name = "BCNN wat",
marker = attr(symbol = 2, color = "rgb(033,115,240)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = (-avg_rate_bcnn .* bayes_acc_mnist_eff),
name = "BCNN acc",
marker = attr(symbol = 0, color = "rgb(151,177,255)", line_width = 1.0),
),
],
Layout(
mode = "overlay",
xaxis_tickangle = -45,
yaxis_title_text = "ΔEfficency",
xaxis_title_text = "Size";
xaxis_range = [-1, 5],
xaxis_type = "category",
),
)
savefig(en_plot, "mnist_eff_exp.png")
en_plot = plot(
[
scatter(
x = ["1", "2", "3", "4", "5"],
y = (-avg_rate_fcnn .* freqs_100_cifar_eff),
name = "LeNet 100",
marker = attr(symbol = 4, color = "rgb(211,120,000)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = (-avg_rate_fcnn .* freqs_est_cifar_eff),
name = "LeNet est",
marker = attr(symbol = 17, color = "rgb(255,170,017)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = (-avg_rate_fcnn .* freqs_wat_cifar_eff),
name = "LeNet wat",
marker = attr(symbol = 2, color = "rgb(255,187,034)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = (-avg_rate_fcnn .* freqs_acc_cifar_eff),
name = "LeNet acc",
marker = attr(symbol = 0, color = "rgb(255,204,051)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = (-avg_rate_bcnn .* bayes_100_cifar_eff),
name = "BCNN 100",
marker = attr(symbol = 4, color = "rgb(055,033,240)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = (-avg_rate_bcnn .* bayes_est_cifar_eff),
name = "BCNN est",
marker = attr(symbol = 17, color = "rgb(033,081,240)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = (-avg_rate_bcnn .* bayes_wat_cifar_eff),
name = "BCNN wat",
marker = attr(symbol = 2, color = "rgb(033,115,240)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = (-avg_rate_bcnn .* bayes_acc_cifar_eff),
name = "BCNN acc",
marker = attr(symbol = 0, color = "rgb(151,177,255)", line_width = 1.0),
),
],
Layout(
mode = "overlay",
xaxis_tickangle = -45,
yaxis_title_text = "ΔEfficency",
xaxis_title_text = "Size";
xaxis_range = [-1, 5],
xaxis_type = "category",
),
)
savefig(en_plot, "cifar_eff_exp.png")

386
efficiency_graphs_sum.jl Executable file
View File

@ -0,0 +1,386 @@
using PlotlyJS
using PlotlyJS: savefig
using Statistics: mean, std
using DataFrames
include("aux_func.jl")
data = load_pickle("efficiency_per_size_sum.pkl");
#all_data_ene["mni"][s]["100"]["bay"]
data_type = ["mni", "cif"]
model_type = ["bay", "frq"]
experiment_type = ["100", "est", "acc", "wat"]
experiment_100=["100","100","100","100","100"]
experiment_est=["est","est","est","est","est"]
experiment_wat=["wat","wat","wat","wat","wat"]
experiment_acc=["acc","acc","acc","acc","acc"]
model_bayes = ["BCNN","BCNN","BCNN","BCNN","BCNN"]
model_lenet = ["LeNet","LeNet","LeNet","LeNet","LeNet"]
data_mnist = ["MNIST","MNIST","MNIST","MNIST","MNIST"]
data_cifar = ["CIFAR","CIFAR","CIFAR","CIFAR","CIFAR"]
sizes = [1,2,3,4,5]
# MNIST 100 efficency
bayes_100_mnist_eff = []
for s = 1:5
push!(
bayes_100_mnist_eff,
data[model_type[1]][data_type[1]][experiment_type[1]][string(s)],
)
end
bayes_100_mnist_eff = DataFrame(Efficiency=bayes_100_mnist_eff,Model=model_bayes,Experiment=experiment_100,Dataset=data_mnist,Size=sizes)
freqs_100_mnist_eff = []
for s = 1:5
push!(
freqs_100_mnist_eff,
data[model_type[2]][data_type[1]][experiment_type[1]][string(s)],
)
end
freqs_100_mnist_eff = DataFrame(Efficiency=freqs_100_mnist_eff,Model=model_lenet,Experiment=experiment_100,Dataset=data_mnist,Size=sizes)
# MNIST est efficency
bayes_est_mnist_eff = []
for s = 1:5
push!(
bayes_est_mnist_eff,
data[model_type[1]][data_type[1]][experiment_type[2]][string(s)],
)
end
bayes_est_mnist_eff = DataFrame(Efficiency=bayes_est_mnist_eff,Model=model_bayes,Experiment=experiment_est,Dataset=data_mnist,Size=sizes)
freqs_est_mnist_eff = []
for s = 1:5
push!(
freqs_est_mnist_eff,
data[model_type[2]][data_type[1]][experiment_type[2]][string(s)],
)
end
freqs_est_mnist_eff = DataFrame(Efficiency=freqs_est_mnist_eff,Model=model_lenet,Experiment=experiment_est,Dataset=data_mnist,Size=sizes)
# MNIST wat efficency
bayes_wat_mnist_eff = []
for s = 1:5
push!(
bayes_wat_mnist_eff,
data[model_type[1]][data_type[1]][experiment_type[4]][string(s)],
)
end
bayes_wat_mnist_eff = DataFrame(Efficiency=bayes_wat_mnist_eff,Model=model_bayes,Experiment=experiment_wat,Dataset=data_mnist,Size=sizes)
freqs_wat_mnist_eff = []
for s = 1:5
push!(
freqs_wat_mnist_eff,
data[model_type[2]][data_type[1]][experiment_type[4]][string(s)],
)
end
freqs_wat_mnist_eff = DataFrame(Efficiency=freqs_wat_mnist_eff,Model=model_lenet,Experiment=experiment_wat,Dataset=data_mnist,Size=sizes)
# MNIST acc efficency
bayes_acc_mnist_eff = []
for s = 1:5
push!(
bayes_acc_mnist_eff,
data[model_type[1]][data_type[1]][experiment_type[3]][string(s)],
)
end
bayes_acc_mnist_eff = DataFrame(Efficiency=bayes_acc_mnist_eff,Model=model_bayes,Experiment=experiment_acc,Dataset=data_mnist,Size=sizes)
freqs_acc_mnist_eff = []
for s = 1:5
push!(
freqs_acc_mnist_eff,
data[model_type[2]][data_type[1]][experiment_type[3]][string(s)],
)
end
freqs_acc_mnist_eff = DataFrame(Efficiency=freqs_acc_mnist_eff,Model=model_lenet,Experiment=experiment_acc,Dataset=data_mnist,Size=sizes)
# CIFAR 100 efficency
bayes_100_cifar_eff = []
for s = 1:5
push!(
bayes_100_cifar_eff,
data[model_type[1]][data_type[2]][experiment_type[1]][string(s)],
)
end
bayes_100_cifar_eff = DataFrame(Efficiency=bayes_100_cifar_eff,Model=model_bayes,Experiment=experiment_100,Dataset=data_cifar,Size=sizes)
#for i = 1:5
# t_std = std(bayes_100_cifar_eff)
# if (bayes_100_cifar_eff[i] > 2 * t_std) || (bayes_100_cifar_eff[i] < 2 * t_std)
# bayes_100_cifar_eff[i] = mean(bayes_100_cifar_eff)
# end
#end
freqs_100_cifar_eff = []
for s = 1:5
push!(
freqs_100_cifar_eff,
data[model_type[2]][data_type[2]][experiment_type[1]][string(s)],
)
end
freqs_100_cifar_eff = DataFrame(Efficiency=freqs_100_cifar_eff,Model=model_lenet,Experiment=experiment_100,Dataset=data_cifar,Size=sizes)
#for i = 1:5
# t_std = std(freqs_100_cifar_eff)
# if (freqs_100_cifar_eff[i] > 2 * t_std) || (freqs_100_cifar_eff[i] < 2 * t_std)
# freqs_100_cifar_eff[i] = mean(freqs_100_cifar_eff)
# end
#end
# CIFAR est efficency
bayes_est_cifar_eff = []
for s = 1:5
push!(
bayes_est_cifar_eff,
data[model_type[1]][data_type[2]][experiment_type[2]][string(s)],
)
end
bayes_est_cifar_eff = DataFrame(Efficiency=bayes_est_cifar_eff,Model=model_bayes,Experiment=experiment_est,Dataset=data_cifar,Size=sizes)
freqs_est_cifar_eff = []
for s = 1:5
push!(
freqs_est_cifar_eff,
data[model_type[2]][data_type[2]][experiment_type[2]][string(s)],
)
end
freqs_est_cifar_eff = DataFrame(Efficiency=freqs_est_cifar_eff,Model=model_lenet,Experiment=experiment_est,Dataset=data_cifar,Size=sizes)
# CIFAR wat efficency
bayes_wat_cifar_eff = []
for s = 1:5
push!(
bayes_wat_cifar_eff,
data[model_type[1]][data_type[2]][experiment_type[4]][string(s)],
)
end
bayes_wat_cifar_eff = DataFrame(Efficiency=bayes_wat_cifar_eff,Model=model_bayes,Experiment=experiment_wat,Dataset=data_cifar,Size=sizes)
freqs_wat_cifar_eff = []
for s = 1:5
push!(
freqs_wat_cifar_eff,
data[model_type[2]][data_type[2]][experiment_type[4]][string(s)],
)
end
freqs_wat_cifar_eff = DataFrame(Efficiency=freqs_wat_cifar_eff,Model=model_lenet,Experiment=experiment_wat,Dataset=data_cifar,Size=sizes)
# CIFAR acc efficency
bayes_acc_cifar_eff = []
for s = 1:5
push!(
bayes_acc_cifar_eff,
data[model_type[1]][data_type[2]][experiment_type[3]][string(s)],
)
end
bayes_acc_cifar_eff = DataFrame(Efficiency=bayes_acc_cifar_eff,Model=model_bayes,Experiment=experiment_acc,Dataset=data_cifar,Size=sizes)
freqs_acc_cifar_eff = []
for s = 1:5
push!(
freqs_acc_cifar_eff,
data[model_type[2]][data_type[2]][experiment_type[3]][string(s)],
)
end
freqs_acc_cifar_eff = DataFrame(Efficiency=freqs_acc_cifar_eff,Model=model_lenet,Experiment=experiment_acc,Dataset=data_cifar,Size=sizes)
mnist_dataframe = vcat(bayes_100_mnist_eff,freqs_100_mnist_eff,bayes_est_mnist_eff,freqs_est_mnist_eff,bayes_acc_mnist_eff,freqs_acc_mnist_eff,bayes_wat_mnist_eff,freqs_wat_mnist_eff)
cifar_dataframe = vcat(bayes_100_cifar_eff,freqs_100_cifar_eff,bayes_est_cifar_eff,freqs_est_cifar_eff,bayes_acc_cifar_eff,freqs_acc_cifar_eff,bayes_wat_cifar_eff,freqs_wat_cifar_eff)
#avg_rate_bcnn = (-8.266684252643054e-5 * 1000)
#avg_rate_fcnn = (0.00022035677966088333 * 1000)
#=
en_plot = plot(
[
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_fcnn .* freqs_100_mnist_eff),
y = freqs_100_mnist_eff.Efficiency,
name = "LeNet 100",
marker = attr(symbol = 4, color = "rgb(211,120,000)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_fcnn .* freqs_est_mnist_eff),
y = freqs_est_mnist_eff.Efficiency,
name = "LeNet est",
marker = attr(symbol = 17, color = "rgb(255,170,017)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_fcnn .* freqs_wat_mnist_eff),
y = freqs_wat_mnist_eff.Efficiency,
name = "LeNet wat",
marker = attr(symbol = 2, color = "rgb(255,187,034)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_fcnn .* freqs_acc_mnist_eff),
y = freqs_acc_mnist_eff.Efficiency,
name = "LeNet acc",
marker = attr(symbol = 0, color = "rgb(255,204,051)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_bcnn .* bayes_100_mnist_eff),
y = bayes_100_mnist_eff.Efficiency,
name = "BCNN 100",
marker = attr(symbol = 4, color = "rgb(055,033,240)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_bcnn .* bayes_est_mnist_eff),
y = bayes_est_mnist_eff.Efficiency,
name = "BCNN est",
marker = attr(symbol = 17, color = "rgb(033,081,240)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_bcnn .* bayes_wat_mnist_eff),
y = bayes_wat_mnist_eff.Efficiency,
name = "BCNN wat",
marker = attr(symbol = 2, color = "rgb(033,115,240)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_bcnn .* bayes_acc_mnist_eff),
y = bayes_acc_mnist_eff.Efficiency,
name = "BCNN acc",
marker = attr(symbol = 0, color = "rgb(151,177,255)", line_width = 1.0),
),
],
Layout(
mode = "overlay",
xaxis_tickangle = -45,
yaxis_title_text = "Efficiency",
xaxis_title_text = "Size";
xaxis_range = [-1, 5],
xaxis_type = "category",
),
)
savefig(en_plot, "mnist_eff_exp_sum.png")
en_plot = plot(
[
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_fcnn .* freqs_100_cifar_eff),
y = freqs_100_cifar_eff.Efficiency,
name = "LeNet 100",
marker = attr(symbol = 4, color = "rgb(211,120,000)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_fcnn .* freqs_est_cifar_eff),
y = freqs_est_cifar_eff.Efficiency,
name = "LeNet est",
marker = attr(symbol = 17, color = "rgb(255,170,017)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_fcnn .* freqs_wat_cifar_eff),
y = freqs_wat_cifar_eff.Efficiency,
name = "LeNet wat",
marker = attr(symbol = 2, color = "rgb(255,187,034)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_fcnn .* freqs_acc_cifar_eff),
y = freqs_acc_cifar_eff.Efficiency,
name = "LeNet acc",
marker = attr(symbol = 0, color = "rgb(255,204,051)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_bcnn .* bayes_100_cifar_eff),
y = bayes_100_cifar_eff.Efficiency,
name = "BCNN 100",
marker = attr(symbol = 4, color = "rgb(055,033,240)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_bcnn .* bayes_est_cifar_eff),
y = bayes_est_cifar_eff.Efficiency,
name = "BCNN est",
marker = attr(symbol = 17, color = "rgb(033,081,240)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_bcnn .* bayes_wat_cifar_eff),
y = bayes_wat_cifar_eff.Efficiency,
name = "BCNN wat",
marker = attr(symbol = 2, color = "rgb(033,115,240)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_bcnn .* bayes_acc_cifar_eff),
y = bayes_acc_cifar_eff.Efficiency,
name = "BCNN acc",
marker = attr(symbol = 0, color = "rgb(151,177,255)", line_width = 1.0),
),
],
Layout(
mode = "overlay",
xaxis_tickangle = -45,
yaxis_title_text = "Efficiency",
xaxis_title_text = "Size";
xaxis_range = [-1, 5],
xaxis_type = "category",
),
)
savefig(en_plot, "cifar_eff_exp_sum.png")
=#
#en_plot = plot(mnist_dataframe, x=:Experiment, y=:Efficiency, boxpoints="all", kind="box")
en_plot = plot(
mnist_dataframe, x=:Experiment, y=:Efficiency,kind="scatter",mode="markers",color=:Model,
quartilemethod="exclusive",
marker=attr(size=:Size, sizeref=maximum(mnist_dataframe.Size) / (10^2), sizemode="area")
#marker=attr(size=:Size, sizeref=0.1, sizemode="area")
)
savefig(en_plot, "mnist_scatter_size.png")
en_plot = plot(
mnist_dataframe, x=:Experiment, y=:Efficiency,kind="box", boxpoints="all",color=:Model,
quartilemethod="exclusive",
marker=attr(size=:Size, sizeref=maximum(mnist_dataframe.Size) / (10^2), sizemode="area")
#marker=attr(size=:Size, sizeref=0.1, sizemode="area")
)
savefig(en_plot, "mnist_box_size.png")
en_plot = plot(
cifar_dataframe, x=:Experiment, y=:Efficiency,kind="scatter",mode="markers",color=:Model,
quartilemethod="exclusive",
marker=attr(size=:Size, sizeref=maximum(mnist_dataframe.Size) / (10^2), sizemode="area")
#marker=attr(size=:Size, sizeref=0.1, sizemode="area")
)
savefig(en_plot, "cifar_scatter_size.png")
en_plot = plot(
cifar_dataframe, x=:Experiment, y=:Efficiency,kind="box", boxpoints="all",color=:Model,
quartilemethod="exclusive",
marker=attr(size=:Size, sizeref=maximum(mnist_dataframe.Size) / (10^2), sizemode="area")
#marker=attr(size=:Size, sizeref=0.1, sizemode="area")
)
savefig(en_plot, "cifar_box_size.png")

1239
energy_compact_processing.jl Executable file
View File

File diff suppressed because it is too large Load Diff

985
energy_measure.jl Executable file
View File

@ -0,0 +1,985 @@
include("aux_func.jl")
using LinearAlgebra
using Statistics
#=
Obtain energy data
=#
#mni_folder_1 = "ini_exp_data/"
mni_folder_100 = "exp_100_epochs/";
cif_folder_100 = "CIFAR_100_epoch/";
mni_folder_acc = "data_bounded/";
cif_folder_acc = "CIFAR_acc_bound/";
mni_folder_wat = "data_budget/";
cif_folder_wat = "CIFAR_energy_bound/";
mni_folder_est = "early_stop_res/";
cif_folder_est = "CIFAR_early_stop/";
bayes_model = "bayes";
freq_model = "freq";
w_type = "watt";
e_type = "exp";
#=
Load GPU data
=#
mni_100_bay_ene = getgpudata(mni_folder_100, bayes_model, w_type, "eo");
mni_100_fre_ene = getgpudata(mni_folder_100, freq_model, w_type, "eo");
cif_100_bay_ene = getgpudata(cif_folder_100, bayes_model, w_type, "eo");
cif_100_fre_ene = getgpudata(cif_folder_100, freq_model, w_type, "eo");
mni_acc_bay_ene = getgpudata(mni_folder_acc, bayes_model, w_type, "eo");
mni_acc_fre_ene = getgpudata(mni_folder_acc, freq_model, w_type, "eo");
cif_acc_bay_ene = getgpudata(cif_folder_acc, bayes_model, w_type, "eo");
cif_acc_fre_ene = getgpudata(cif_folder_acc, freq_model, w_type, "eo");
mni_wat_bay_ene = getgpudata(mni_folder_wat, bayes_model, w_type, "eo");
mni_wat_fre_ene = getgpudata(mni_folder_wat, freq_model, w_type, "eo");
cif_wat_bay_ene = getgpudata(cif_folder_wat, bayes_model, w_type, "eo");
cif_wat_fre_ene = getgpudata(cif_folder_wat, freq_model, w_type, "eo");
mni_est_bay_ene = getgpudata(mni_folder_est, bayes_model, w_type, "eo");
mni_est_fre_ene = getgpudata(mni_folder_est, freq_model, w_type, "eo");
cif_est_bay_ene = getgpudata(cif_folder_est, bayes_model, w_type, "eo");
cif_est_fre_ene = getgpudata(cif_folder_est, freq_model, w_type, "eo");
#=
Define CPU paths
=#
cif = "cifar";
mni = "mnist";
bay = "bayes";
wat = "cpu_watts";
frq = "freq";
ram = "ram_use";
_100 = "100";
acc = "acc";
es = "es";
wbud = "wbud";
#=
Load CPU data
=#
bay_cif_100_ene = readcpudata(cif, bay, wat, _100);
bay_mni_100_ene = readcpudata(mni, bay, wat, _100);
frq_cif_100_ene = readcpudata(cif, frq, wat, _100);
frq_mni_100_ene = readcpudata(mni, frq, wat, _100);
bay_cif_acc_ene = readcpudata(cif, bay, wat, acc);
bay_mni_acc_ene = readcpudata(mni, bay, wat, acc);
frq_cif_acc_ene = readcpudata(cif, frq, wat, acc);
frq_mni_acc_ene = readcpudata(mni, frq, wat, acc);
bay_cif_wbu_ene = readcpudata(cif, bay, wat, wbud);
bay_mni_wbu_ene = readcpudata(mni, bay, wat, wbud);
frq_cif_wbu_ene = readcpudata(cif, frq, wat, wbud);
frq_mni_wbu_ene = readcpudata(mni, frq, wat, wbud);
bay_mni_est_ene = readcpudata(mni, bay, wat, es);
bay_cif_est_ene = readcpudata(cif, bay, wat, es);
frq_cif_est_ene = readcpudata(cif, frq, wat, es);
frq_mni_est_ene = readcpudata(mni, frq, wat, es);
for s = 1:5
bay_cif_acc_ene[s] = round.(getcpuwatt(bay_cif_acc_ene[s]))
bay_cif_est_ene[s] = round.(getcpuwatt(bay_cif_est_ene[s]))
bay_cif_wbu_ene[s] = round.(getcpuwatt(bay_cif_wbu_ene[s]))
bay_cif_100_ene[s] = round.(getcpuwatt(bay_cif_100_ene[s]))
bay_mni_acc_ene[s] = round.(getcpuwatt(bay_mni_acc_ene[s]))
bay_mni_est_ene[s] = round.(getcpuwatt(bay_mni_est_ene[s]))
bay_mni_wbu_ene[s] = round.(getcpuwatt(bay_mni_wbu_ene[s]))
bay_mni_100_ene[s] = round.(getcpuwatt(bay_mni_100_ene[s]))
frq_cif_acc_ene[s] = round.(getcpuwatt(frq_cif_acc_ene[s]))
frq_cif_est_ene[s] = round.(getcpuwatt(frq_cif_est_ene[s]))
frq_cif_wbu_ene[s] = round.(getcpuwatt(frq_cif_wbu_ene[s]))
frq_cif_100_ene[s] = round.(getcpuwatt(frq_cif_100_ene[s]))
frq_mni_acc_ene[s] = round.(getcpuwatt(frq_mni_acc_ene[s]))
frq_mni_est_ene[s] = round.(getcpuwatt(frq_mni_est_ene[s]))
frq_mni_wbu_ene[s] = round.(getcpuwatt(frq_mni_wbu_ene[s]))
frq_mni_100_ene[s] = round.(getcpuwatt(frq_mni_100_ene[s]))
end
for s = 1:5
mni_100_bay_ene[s]["Ene"] = vcat(mni_100_bay_ene[s]["Ene"], bay_mni_100_ene[s])
mni_100_fre_ene[s]["Ene"] = vcat(mni_100_fre_ene[s]["Ene"], frq_mni_100_ene[s])
cif_100_bay_ene[s]["Ene"] = vcat(cif_100_bay_ene[s]["Ene"], bay_cif_100_ene[s])
cif_100_fre_ene[s]["Ene"] = vcat(cif_100_fre_ene[s]["Ene"], frq_cif_100_ene[s])
mni_acc_bay_ene[s]["Ene"] = vcat(mni_acc_bay_ene[s]["Ene"], bay_mni_acc_ene[s])
mni_acc_fre_ene[s]["Ene"] = vcat(mni_acc_fre_ene[s]["Ene"], frq_mni_acc_ene[s])
cif_acc_bay_ene[s]["Ene"] = vcat(cif_acc_bay_ene[s]["Ene"], bay_cif_acc_ene[s])
cif_acc_fre_ene[s]["Ene"] = vcat(cif_acc_fre_ene[s]["Ene"], frq_cif_acc_ene[s])
mni_wat_bay_ene[s]["Ene"] = vcat(mni_wat_bay_ene[s]["Ene"], bay_mni_wbu_ene[s])
mni_wat_fre_ene[s]["Ene"] = vcat(mni_wat_fre_ene[s]["Ene"], frq_mni_wbu_ene[s])
cif_wat_bay_ene[s]["Ene"] = vcat(cif_wat_bay_ene[s]["Ene"], bay_cif_wbu_ene[s])
cif_wat_fre_ene[s]["Ene"] = vcat(cif_wat_fre_ene[s]["Ene"], frq_cif_wbu_ene[s])
mni_est_bay_ene[s]["Ene"] = vcat(mni_est_bay_ene[s]["Ene"], bay_mni_est_ene[s])
mni_est_fre_ene[s]["Ene"] = vcat(mni_est_fre_ene[s]["Ene"], frq_mni_est_ene[s])
cif_est_bay_ene[s]["Ene"] = vcat(cif_est_bay_ene[s]["Ene"], bay_cif_est_ene[s])
cif_est_fre_ene[s]["Ene"] = vcat(cif_est_fre_ene[s]["Ene"], frq_cif_est_ene[s])
end
all_data_ene = Dict(
"mni" => Dict(
1 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
2 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
3 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
4 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
5 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
),
"cif" => Dict(
1 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
2 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
3 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
4 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
5 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
),
)
for s = 1:5
all_data_ene["mni"][s]["100"]["bay"] = sum(mni_100_bay_ene[s]["Ene"])
println(
"MNIST Bayes Mean energy size $(s) 100: $(all_data_ene["mni"][s]["100"]["bay"])",
)
all_data_ene["mni"][s]["100"]["frq"] = sum(mni_100_fre_ene[s]["Ene"])
println("MNIST Freq Mean energy size $(s) 100: $(all_data_ene["mni"][s]["100"]["frq"])")
all_data_ene["cif"][s]["100"]["bay"] = sum(cif_100_bay_ene[s]["Ene"])
println(
"CIFAR Bayes Mean energy size $(s) 100: $(all_data_ene["cif"][s]["100"]["bay"])",
)
all_data_ene["cif"][s]["100"]["frq"] = sum(cif_100_fre_ene[s]["Ene"])
println("CIFAR Freq Mean energy size $(s) 100: $(all_data_ene["cif"][s]["100"]["frq"])")
all_data_ene["mni"][s]["acc"]["bay"] = sum(mni_acc_bay_ene[s]["Ene"])
println(
"MNIST Bayes Mean energy size $(s) Acc: $(all_data_ene["mni"][s]["acc"]["bay"])",
)
all_data_ene["mni"][s]["acc"]["frq"] = sum(mni_acc_fre_ene[s]["Ene"])
println("MNIST Freq Mean energy size $(s) Acc: $(all_data_ene["mni"][s]["acc"]["frq"])")
all_data_ene["cif"][s]["acc"]["bay"] = sum(cif_acc_bay_ene[s]["Ene"])
println(
"CIFAR Bayes Mean energy size $(s) Acc: $(all_data_ene["cif"][s]["acc"]["bay"])",
)
all_data_ene["cif"][s]["acc"]["frq"] = sum(cif_acc_fre_ene[s]["Ene"])
println("CIFAR Freq Mean energy size $(s) Acc: $(all_data_ene["cif"][s]["acc"]["frq"])")
all_data_ene["mni"][s]["wat"]["bay"] = sum(mni_wat_bay_ene[s]["Ene"])
println(
"MNIST Bayes Mean energy size $(s) Wat: $(all_data_ene["mni"][s]["wat"]["bay"])",
)
all_data_ene["mni"][s]["wat"]["frq"] = sum(mni_wat_fre_ene[s]["Ene"])
println("MNIST Freq Mean energy size $(s) Wat: $(all_data_ene["mni"][s]["wat"]["frq"])")
all_data_ene["cif"][s]["wat"]["bay"] = sum(cif_wat_bay_ene[s]["Ene"])
println(
"CIFAR Bayes Mean energy size $(s) Wat: $(all_data_ene["cif"][s]["wat"]["bay"])",
)
all_data_ene["cif"][s]["wat"]["frq"] = sum(cif_wat_fre_ene[s]["Ene"])
println("CIFAR Freq Mean energy size $(s) Wat: $(all_data_ene["cif"][s]["wat"]["frq"])")
all_data_ene["mni"][s]["est"]["bay"] = sum(mni_est_bay_ene[s]["Ene"])
println(
"MNIST Bayes Mean energy size $(s) Est: $(all_data_ene["mni"][s]["est"]["bay"])",
)
all_data_ene["mni"][s]["est"]["frq"] = sum(mni_est_fre_ene[s]["Ene"])
println("MNIST Freq Mean energy size $(s) Est: $(all_data_ene["mni"][s]["est"]["frq"])")
all_data_ene["cif"][s]["est"]["bay"] = sum(cif_est_bay_ene[s]["Ene"])
println(
"CIFAR Bayes Mean energy size $(s) Est: $(all_data_ene["cif"][s]["est"]["bay"])",
)
all_data_ene["cif"][s]["est"]["frq"] = sum(cif_est_fre_ene[s]["Ene"])
println("CIFAR Freq Mean energy size $(s) Est: $(all_data_ene["cif"][s]["est"]["frq"])")
end
#=
Load Accuracy data
=#
mni_100_bay_exp = getgpudata(mni_folder_100, bayes_model, e_type, "eo");
mni_100_fre_exp = getgpudata(mni_folder_100, freq_model, e_type, "eo");
cif_100_bay_exp = getgpudata(cif_folder_100, bayes_model, e_type, "eo");
cif_100_fre_exp = getgpudata(cif_folder_100, freq_model, e_type, "eo");
mni_acc_bay_exp = getgpudata(mni_folder_acc, bayes_model, e_type, "eo");
mni_acc_fre_exp = getgpudata(mni_folder_acc, freq_model, e_type, "eo");
cif_acc_bay_exp = getgpudata(cif_folder_acc, bayes_model, e_type, "eo");
cif_acc_fre_exp = getgpudata(cif_folder_acc, freq_model, e_type, "eo");
mni_wat_bay_exp = getgpudata(mni_folder_wat, bayes_model, e_type, "eo");
mni_wat_fre_exp = getgpudata(mni_folder_wat, freq_model, e_type, "eo");
cif_wat_bay_exp = getgpudata(cif_folder_wat, bayes_model, e_type, "eo");
cif_wat_fre_exp = getgpudata(cif_folder_wat, freq_model, e_type, "eo");
mni_est_bay_exp = getgpudata(mni_folder_est, bayes_model, e_type, "eo");
mni_est_fre_exp = getgpudata(mni_folder_est, freq_model, e_type, "eo");
cif_est_bay_exp = getgpudata(cif_folder_est, bayes_model, e_type, "eo");
cif_est_fre_exp = getgpudata(cif_folder_est, freq_model, e_type, "eo");
for i = 1:5
mni_100_bay_exp[i] = expdatatodict(mni_100_bay_exp[i])
mni_100_fre_exp[i] = expdatatodict(mni_100_fre_exp[i])
cif_100_bay_exp[i] = expdatatodict(cif_100_bay_exp[i])
cif_100_fre_exp[i] = expdatatodict(cif_100_fre_exp[i])
mni_acc_bay_exp[i] = expdatatodict(mni_acc_bay_exp[i])
mni_acc_fre_exp[i] = expdatatodict(mni_acc_fre_exp[i])
cif_acc_bay_exp[i] = expdatatodict(cif_acc_bay_exp[i])
cif_acc_fre_exp[i] = expdatatodict(cif_acc_fre_exp[i])
mni_wat_bay_exp[i] = expdatatodict(mni_wat_bay_exp[i])
mni_wat_fre_exp[i] = expdatatodict(mni_wat_fre_exp[i])
cif_wat_bay_exp[i] = expdatatodict(cif_wat_bay_exp[i])
cif_wat_fre_exp[i] = expdatatodict(cif_wat_fre_exp[i])
mni_est_bay_exp[i] = expdatatodict(mni_est_bay_exp[i])
mni_est_fre_exp[i] = expdatatodict(mni_est_fre_exp[i])
cif_est_bay_exp[i] = expdatatodict(cif_est_bay_exp[i])
cif_est_fre_exp[i] = expdatatodict(cif_est_fre_exp[i])
end
all_data_exp = Dict(
"mni" => Dict(
1 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
2 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
3 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
4 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
5 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
),
"cif" => Dict(
1 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
2 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
3 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
4 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
5 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
),
)
for s = 1:5
all_data_exp["mni"][s]["100"]["bay"] =
(mean(mni_100_bay_exp[s]["acc"]) + mean(mni_100_bay_exp[s]["pre"])) / 2
println(
"MNIST Bayes Mean accuracy size $(s) 100: $(all_data_exp["mni"][s]["100"]["bay"])",
)
all_data_exp["mni"][s]["100"]["frq"] =
(mean(mni_100_fre_exp[s]["acc"]) + mean(mni_100_fre_exp[s]["pre"])) / 2
println(
"MNIST Freq Mean accuracy size $(s) 100: $(all_data_exp["mni"][s]["100"]["frq"])",
)
all_data_exp["cif"][s]["100"]["bay"] =
(mean(cif_100_bay_exp[s]["acc"]) + mean(cif_100_bay_exp[s]["pre"])) / 2
println(
"CIFAR Bayes Mean accuracy size $(s) 100: $(all_data_exp["cif"][s]["100"]["bay"])",
)
all_data_exp["cif"][s]["100"]["frq"] =
(mean(cif_100_fre_exp[s]["acc"]) + mean(cif_100_fre_exp[s]["pre"])) / 2
println(
"CIFAR Freq Mean accuracy size $(s) 100: $(all_data_exp["cif"][s]["100"]["frq"])",
)
all_data_exp["mni"][s]["acc"]["bay"] =
(mean(mni_acc_bay_exp[s]["acc"]) + mean(mni_acc_bay_exp[s]["pre"])) / 2
println(
"MNIST Bayes Mean accuracy size $(s) Acc: $(all_data_exp["mni"][s]["acc"]["bay"])",
)
all_data_exp["mni"][s]["acc"]["frq"] =
(mean(mni_acc_fre_exp[s]["acc"]) + mean(mni_acc_fre_exp[s]["pre"])) / 2
println(
"MNIST Freq Mean accuracy size $(s) Acc: $(all_data_exp["mni"][s]["acc"]["frq"])",
)
all_data_exp["cif"][s]["acc"]["bay"] =
(mean(cif_acc_bay_exp[s]["acc"]) + mean(cif_acc_bay_exp[s]["pre"])) / 2
println(
"CIFAR Bayes Mean accuracy size $(s) Acc: $(all_data_exp["cif"][s]["acc"]["bay"])",
)
all_data_exp["cif"][s]["acc"]["frq"] =
(mean(cif_acc_fre_exp[s]["acc"]) + mean(cif_acc_fre_exp[s]["pre"])) / 2
println(
"CIFAR Freq Mean accuracy size $(s) Acc: $(all_data_exp["cif"][s]["acc"]["frq"])",
)
all_data_exp["mni"][s]["wat"]["bay"] =
(mean(mni_wat_bay_exp[s]["acc"]) + mean(mni_wat_bay_exp[s]["pre"])) / 2
println(
"MNIST Bayes Mean accuracy size $(s) Wat: $(all_data_exp["mni"][s]["wat"]["bay"])",
)
all_data_exp["mni"][s]["wat"]["frq"] =
(mean(mni_wat_fre_exp[s]["acc"]) + mean(mni_wat_fre_exp[s]["pre"])) / 2
println(
"MNIST Freq Mean accuracy size $(s) Wat: $(all_data_exp["mni"][s]["wat"]["frq"])",
)
all_data_exp["cif"][s]["wat"]["bay"] =
(mean(cif_wat_bay_exp[s]["acc"]) + mean(cif_wat_bay_exp[s]["pre"])) / 2
println(
"CIFAR Bayes Mean accuracy size $(s) Wat: $(all_data_exp["cif"][s]["wat"]["bay"])",
)
all_data_exp["cif"][s]["wat"]["frq"] =
(mean(cif_wat_fre_exp[s]["acc"]) + mean(cif_wat_fre_exp[s]["pre"])) / 2
println(
"CIFAR Freq Mean accuracy size $(s) Wat: $(all_data_exp["cif"][s]["wat"]["frq"])",
)
all_data_exp["mni"][s]["est"]["bay"] =
(mean(mni_est_bay_exp[s]["acc"]) + mean(mni_est_bay_exp[s]["pre"])) / 2
println(
"MNIST Bayes Mean accuracy size $(s) Est: $(all_data_exp["mni"][s]["est"]["bay"])",
)
all_data_exp["mni"][s]["est"]["frq"] =
(mean(mni_est_fre_exp[s]["acc"]) + mean(mni_est_fre_exp[s]["pre"])) / 2
println(
"MNIST Freq Mean accuracy size $(s) Est: $(all_data_exp["mni"][s]["est"]["frq"])",
)
all_data_exp["cif"][s]["est"]["bay"] =
(mean(cif_est_bay_exp[s]["acc"]) + mean(cif_est_bay_exp[s]["pre"])) / 2
println(
"CIFAR Bayes Mean accuracy size $(s) Est: $(all_data_exp["cif"][s]["est"]["bay"])",
)
all_data_exp["cif"][s]["est"]["frq"] =
(mean(cif_est_fre_exp[s]["acc"]) + mean(cif_est_fre_exp[s]["pre"])) / 2
println(
"CIFAR Freq Mean accuracy size $(s) Est: $(all_data_exp["cif"][s]["est"]["frq"])",
)
end
#=
# Per size efficiency
=#
efficiency_per_size = Dict(
"frq" => Dict(
"mni" => Dict(
"100" => Dict("1" => 0.0, "2" => 0.0, "3" => 0.0, "4" => 0.0, "5" => 0.0),
"est" => Dict("1" => 0.0, "2" => 0.0, "3" => 0.0, "4" => 0.0, "5" => 0.0),
"acc" => Dict("1" => 0.0, "2" => 0.0, "3" => 0.0, "4" => 0.0, "5" => 0.0),
"wat" => Dict("1" => 0.0, "2" => 0.0, "3" => 0.0, "4" => 0.0, "5" => 0.0),
),
"cif" => Dict(
"100" => Dict("1" => 0.0, "2" => 0.0, "3" => 0.0, "4" => 0.0, "5" => 0.0),
"est" => Dict("1" => 0.0, "2" => 0.0, "3" => 0.0, "4" => 0.0, "5" => 0.0),
"acc" => Dict("1" => 0.0, "2" => 0.0, "3" => 0.0, "4" => 0.0, "5" => 0.0),
"wat" => Dict("1" => 0.0, "2" => 0.0, "3" => 0.0, "4" => 0.0, "5" => 0.0),
),
),
"bay" => Dict(
"mni" => Dict(
"100" => Dict("1" => 0.0, "2" => 0.0, "3" => 0.0, "4" => 0.0, "5" => 0.0),
"est" => Dict("1" => 0.0, "2" => 0.0, "3" => 0.0, "4" => 0.0, "5" => 0.0),
"acc" => Dict("1" => 0.0, "2" => 0.0, "3" => 0.0, "4" => 0.0, "5" => 0.0),
"wat" => Dict("1" => 0.0, "2" => 0.0, "3" => 0.0, "4" => 0.0, "5" => 0.0),
),
"cif" => Dict(
"100" => Dict("1" => 0.0, "2" => 0.0, "3" => 0.0, "4" => 0.0, "5" => 0.0),
"est" => Dict("1" => 0.0, "2" => 0.0, "3" => 0.0, "4" => 0.0, "5" => 0.0),
"acc" => Dict("1" => 0.0, "2" => 0.0, "3" => 0.0, "4" => 0.0, "5" => 0.0),
"wat" => Dict("1" => 0.0, "2" => 0.0, "3" => 0.0, "4" => 0.0, "5" => 0.0),
),
),
)
data_type = ["mni", "cif"]
model_type = ["bay", "frq"]
experiment_type = ["100", "est", "acc", "wat"]
what = collect(Iterators.product(model_type, data_type, experiment_type, 1:5))
for t in what
#println("model: $(t[1]), data: $(t[2]), experiment: $(t[3]), size: $(t[4])")
#println(all_data_exp[t[2]][t[4]][t[3]][t[1]]/all_data_ene[t[2]][t[4]][t[3]][t[1]] * 100)
efficiency_per_size[t[1]][t[2]][t[3]][string(t[4])] =
all_data_exp[t[2]][t[4]][t[3]][t[1]] / all_data_ene[t[2]][t[4]][t[3]][t[1]] * 100
end
save_pickle("efficiency_per_size.pkl", efficiency_per_size)
save_pickle("energy_complete.pkl", all_data_ene)
#=
# Accuracy means
=#
comp_exp = Dict(
"mni" => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
"cif" => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
)
comp_exp["mni"]["100"]["frq"] =
(
all_data_exp["mni"][1]["100"]["frq"] +
all_data_exp["mni"][2]["100"]["frq"] +
all_data_exp["mni"][3]["100"]["frq"] +
all_data_exp["mni"][4]["100"]["frq"] +
all_data_exp["mni"][5]["100"]["frq"]
) / 5
comp_exp["mni"]["100"]["bay"] =
(
all_data_exp["mni"][1]["100"]["bay"] +
all_data_exp["mni"][2]["100"]["bay"] +
all_data_exp["mni"][3]["100"]["bay"] +
all_data_exp["mni"][4]["100"]["bay"] +
all_data_exp["mni"][5]["100"]["bay"]
) / 5
comp_exp["cif"]["100"]["frq"] =
(
all_data_exp["cif"][1]["100"]["frq"] +
all_data_exp["cif"][2]["100"]["frq"] +
all_data_exp["cif"][3]["100"]["frq"] +
all_data_exp["cif"][4]["100"]["frq"] +
all_data_exp["cif"][5]["100"]["frq"]
) / 5
comp_exp["cif"]["100"]["bay"] =
(
all_data_exp["cif"][1]["100"]["bay"] +
all_data_exp["cif"][2]["100"]["bay"] +
all_data_exp["cif"][3]["100"]["bay"] +
all_data_exp["cif"][4]["100"]["bay"] +
all_data_exp["cif"][5]["100"]["bay"]
) / 5
comp_exp["mni"]["acc"]["frq"] =
(
all_data_exp["mni"][1]["acc"]["frq"] +
all_data_exp["mni"][2]["acc"]["frq"] +
all_data_exp["mni"][3]["acc"]["frq"] +
all_data_exp["mni"][4]["acc"]["frq"] +
all_data_exp["mni"][5]["acc"]["frq"]
) / 5
comp_exp["mni"]["acc"]["bay"] =
(
all_data_exp["mni"][1]["acc"]["bay"] +
all_data_exp["mni"][2]["acc"]["bay"] +
all_data_exp["mni"][3]["acc"]["bay"] +
all_data_exp["mni"][4]["acc"]["bay"] +
all_data_exp["mni"][5]["acc"]["bay"]
) / 5
comp_exp["cif"]["acc"]["frq"] =
(
all_data_exp["cif"][1]["acc"]["frq"] +
all_data_exp["cif"][2]["acc"]["frq"] +
all_data_exp["cif"][3]["acc"]["frq"] +
all_data_exp["cif"][4]["acc"]["frq"] +
all_data_exp["cif"][5]["acc"]["frq"]
) / 5
comp_exp["cif"]["acc"]["bay"] =
(
all_data_exp["cif"][1]["acc"]["bay"] +
all_data_exp["cif"][2]["acc"]["bay"] +
all_data_exp["cif"][3]["acc"]["bay"] +
all_data_exp["cif"][4]["acc"]["bay"] +
all_data_exp["cif"][5]["acc"]["bay"]
) / 5
comp_exp["mni"]["wat"]["frq"] =
(
all_data_exp["mni"][1]["wat"]["frq"] +
all_data_exp["mni"][2]["wat"]["frq"] +
all_data_exp["mni"][3]["wat"]["frq"] +
all_data_exp["mni"][4]["wat"]["frq"] +
all_data_exp["mni"][5]["wat"]["frq"]
) / 5
comp_exp["mni"]["wat"]["bay"] =
(
all_data_exp["mni"][1]["wat"]["bay"] +
all_data_exp["mni"][2]["wat"]["bay"] +
all_data_exp["mni"][3]["wat"]["bay"] +
all_data_exp["mni"][4]["wat"]["bay"] +
all_data_exp["mni"][5]["wat"]["bay"]
) / 5
comp_exp["cif"]["wat"]["frq"] =
(
all_data_exp["cif"][1]["wat"]["frq"] +
all_data_exp["cif"][2]["wat"]["frq"] +
all_data_exp["cif"][3]["wat"]["frq"] +
all_data_exp["cif"][4]["wat"]["frq"] +
all_data_exp["cif"][5]["wat"]["frq"]
) / 5
comp_exp["cif"]["wat"]["bay"] =
(
all_data_exp["cif"][1]["wat"]["bay"] +
all_data_exp["cif"][2]["wat"]["bay"] +
all_data_exp["cif"][3]["wat"]["bay"] +
all_data_exp["cif"][4]["wat"]["bay"] +
all_data_exp["cif"][5]["wat"]["bay"]
) / 5
comp_exp["mni"]["est"]["frq"] =
(
all_data_exp["mni"][1]["est"]["frq"] +
all_data_exp["mni"][2]["est"]["frq"] +
all_data_exp["mni"][3]["est"]["frq"] +
all_data_exp["mni"][4]["est"]["frq"] +
all_data_exp["mni"][5]["est"]["frq"]
) / 5
comp_exp["mni"]["est"]["bay"] =
(
all_data_exp["mni"][1]["est"]["bay"] +
all_data_exp["mni"][2]["est"]["bay"] +
all_data_exp["mni"][3]["est"]["bay"] +
all_data_exp["mni"][4]["est"]["bay"] +
all_data_exp["mni"][5]["est"]["bay"]
) / 5
comp_exp["cif"]["est"]["frq"] =
(
all_data_exp["cif"][1]["est"]["frq"] +
all_data_exp["cif"][2]["est"]["frq"] +
all_data_exp["cif"][3]["est"]["frq"] +
all_data_exp["cif"][4]["est"]["frq"] +
all_data_exp["cif"][5]["est"]["frq"]
) / 5
comp_exp["cif"]["est"]["bay"] =
(
all_data_exp["cif"][1]["est"]["bay"] +
all_data_exp["cif"][2]["est"]["bay"] +
all_data_exp["cif"][3]["est"]["bay"] +
all_data_exp["cif"][4]["est"]["bay"] +
all_data_exp["cif"][5]["est"]["bay"]
) / 5
#=
# Energy means
=#
comp_ene = Dict(
"mni" => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
"cif" => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
)
comp_ene["mni"]["100"]["frq"] =
(
all_data_ene["mni"][1]["100"]["frq"] +
all_data_ene["mni"][2]["100"]["frq"] +
all_data_ene["mni"][3]["100"]["frq"] +
all_data_ene["mni"][4]["100"]["frq"] +
all_data_ene["mni"][5]["100"]["frq"]
) / 5
comp_ene["mni"]["100"]["bay"] =
(
all_data_ene["mni"][1]["100"]["bay"] +
all_data_ene["mni"][2]["100"]["bay"] +
all_data_ene["mni"][3]["100"]["bay"] +
all_data_ene["mni"][4]["100"]["bay"] +
all_data_ene["mni"][5]["100"]["bay"]
) / 5
comp_ene["cif"]["100"]["frq"] =
(
all_data_ene["cif"][1]["100"]["frq"] +
all_data_ene["cif"][2]["100"]["frq"] +
all_data_ene["cif"][3]["100"]["frq"] +
all_data_ene["cif"][4]["100"]["frq"] +
all_data_ene["cif"][5]["100"]["frq"]
) / 5
comp_ene["cif"]["100"]["bay"] =
(
all_data_ene["cif"][1]["100"]["bay"] +
all_data_ene["cif"][2]["100"]["bay"] +
all_data_ene["cif"][3]["100"]["bay"] +
all_data_ene["cif"][4]["100"]["bay"] +
all_data_ene["cif"][5]["100"]["bay"]
) / 5
comp_ene["mni"]["acc"]["frq"] =
(
all_data_ene["mni"][1]["acc"]["frq"] +
all_data_ene["mni"][2]["acc"]["frq"] +
all_data_ene["mni"][3]["acc"]["frq"] +
all_data_ene["mni"][4]["acc"]["frq"] +
all_data_ene["mni"][5]["acc"]["frq"]
) / 5
comp_ene["mni"]["acc"]["bay"] =
(
all_data_ene["mni"][1]["acc"]["bay"] +
all_data_ene["mni"][2]["acc"]["bay"] +
all_data_ene["mni"][3]["acc"]["bay"] +
all_data_ene["mni"][4]["acc"]["bay"] +
all_data_ene["mni"][5]["acc"]["bay"]
) / 5
comp_ene["cif"]["acc"]["frq"] =
(
all_data_ene["cif"][1]["acc"]["frq"] +
all_data_ene["cif"][2]["acc"]["frq"] +
all_data_ene["cif"][3]["acc"]["frq"] +
all_data_ene["cif"][4]["acc"]["frq"] +
all_data_ene["cif"][5]["acc"]["frq"]
) / 5
comp_ene["cif"]["acc"]["bay"] =
(
all_data_ene["cif"][1]["acc"]["bay"] +
all_data_ene["cif"][2]["acc"]["bay"] +
all_data_ene["cif"][3]["acc"]["bay"] +
all_data_ene["cif"][4]["acc"]["bay"] +
all_data_ene["cif"][5]["acc"]["bay"]
) / 5
comp_ene["mni"]["wat"]["frq"] =
(
all_data_ene["mni"][1]["wat"]["frq"] +
all_data_ene["mni"][2]["wat"]["frq"] +
all_data_ene["mni"][3]["wat"]["frq"] +
all_data_ene["mni"][4]["wat"]["frq"] +
all_data_ene["mni"][5]["wat"]["frq"]
) / 5
comp_ene["mni"]["wat"]["bay"] =
(
all_data_ene["mni"][1]["wat"]["bay"] +
all_data_ene["mni"][2]["wat"]["bay"] +
all_data_ene["mni"][3]["wat"]["bay"] +
all_data_ene["mni"][4]["wat"]["bay"] +
all_data_ene["mni"][5]["wat"]["bay"]
) / 5
comp_ene["cif"]["wat"]["frq"] =
(
all_data_ene["cif"][1]["wat"]["frq"] +
all_data_ene["cif"][2]["wat"]["frq"] +
all_data_ene["cif"][3]["wat"]["frq"] +
all_data_ene["cif"][4]["wat"]["frq"] +
all_data_ene["cif"][5]["wat"]["frq"]
) / 5
comp_ene["cif"]["wat"]["bay"] =
(
all_data_ene["cif"][1]["wat"]["bay"] +
all_data_ene["cif"][2]["wat"]["bay"] +
all_data_ene["cif"][3]["wat"]["bay"] +
all_data_ene["cif"][4]["wat"]["bay"] +
all_data_ene["cif"][5]["wat"]["bay"]
) / 5
comp_ene["mni"]["est"]["frq"] =
(
all_data_ene["mni"][1]["est"]["frq"] +
all_data_ene["mni"][2]["est"]["frq"] +
all_data_ene["mni"][3]["est"]["frq"] +
all_data_ene["mni"][4]["est"]["frq"] +
all_data_ene["mni"][5]["est"]["frq"]
) / 5
comp_ene["mni"]["est"]["bay"] =
(
all_data_ene["mni"][1]["est"]["bay"] +
all_data_ene["mni"][2]["est"]["bay"] +
all_data_ene["mni"][3]["est"]["bay"] +
all_data_ene["mni"][4]["est"]["bay"] +
all_data_ene["mni"][5]["est"]["bay"]
) / 5
comp_ene["cif"]["est"]["frq"] =
(
all_data_ene["cif"][1]["est"]["frq"] +
all_data_ene["cif"][2]["est"]["frq"] +
all_data_ene["cif"][3]["est"]["frq"] +
all_data_ene["cif"][4]["est"]["frq"] +
all_data_ene["cif"][5]["est"]["frq"]
) / 5
comp_ene["cif"]["est"]["bay"] =
(
all_data_ene["cif"][1]["est"]["bay"] +
all_data_ene["cif"][2]["est"]["bay"] +
all_data_ene["cif"][3]["est"]["bay"] +
all_data_ene["cif"][4]["est"]["bay"] +
all_data_ene["cif"][5]["est"]["bay"]
) / 5
#=
Final Metric per experiment
=#
println(
"MNIST LeNet 100, Accuracy: $(comp_exp["mni"]["100"]["frq"]), Energy: $(comp_ene["mni"]["100"]["frq"]), Ratio: $((comp_exp["mni"]["100"]["frq"] / comp_ene["mni"]["100"]["frq"]) * 100)",
)
println(
"MNIST Bayes 100, Accuracy: $(comp_exp["mni"]["100"]["bay"]), Energy: $(comp_ene["mni"]["100"]["bay"]), Ratio: $((comp_exp["mni"]["100"]["bay"] / comp_ene["mni"]["100"]["bay"]) * 100)",
)
println(
"CIFAR LeNet 100, Accuracy: $(comp_exp["cif"]["100"]["frq"]), Energy: $(comp_ene["cif"]["100"]["frq"]), Ratio: $((comp_exp["cif"]["100"]["frq"] / comp_ene["cif"]["100"]["frq"]) * 100)",
)
println(
"CIFAR Bayes 100, Accuracy: $(comp_exp["cif"]["100"]["bay"]), Energy: $(comp_ene["cif"]["100"]["bay"]), Ratio: $((comp_exp["cif"]["100"]["bay"] / comp_ene["cif"]["100"]["bay"]) * 100)",
)
println(
"MNIST LeNet Acc, Accuracy: $(comp_exp["mni"]["acc"]["frq"]), Energy: $(comp_ene["mni"]["acc"]["frq"]), Ratio: $((comp_exp["mni"]["acc"]["frq"] / comp_ene["mni"]["acc"]["frq"]) * 100)",
)
println(
"MNIST Bayes Acc, Accuracy: $(comp_exp["mni"]["acc"]["bay"]), Energy: $(comp_ene["mni"]["acc"]["bay"]), Ratio: $((comp_exp["mni"]["acc"]["bay"] / comp_ene["mni"]["acc"]["bay"]) * 100)",
)
println(
"CIFAR LeNet Acc, Accuracy: $(comp_exp["cif"]["acc"]["frq"]), Energy: $(comp_ene["cif"]["acc"]["frq"]), Ratio: $((comp_exp["cif"]["acc"]["frq"] / comp_ene["cif"]["acc"]["frq"]) * 100)",
)
println(
"CIFAR Bayes Acc, Accuracy: $(comp_exp["cif"]["acc"]["bay"]), Energy: $(comp_ene["cif"]["acc"]["bay"]), Ratio: $((comp_exp["cif"]["acc"]["bay"] / comp_ene["cif"]["acc"]["bay"]) * 100)",
)
println(
"MNIST LeNet Wat, Accuracy: $(comp_exp["mni"]["wat"]["frq"]), Energy: $(comp_ene["mni"]["wat"]["frq"]), Ratio: $((comp_exp["mni"]["wat"]["frq"] / comp_ene["mni"]["wat"]["frq"]) * 100)",
)
println(
"MNIST Bayes Wat, Accuracy: $(comp_exp["mni"]["wat"]["bay"]), Energy: $(comp_ene["mni"]["wat"]["bay"]), Ratio: $((comp_exp["mni"]["wat"]["bay"] / comp_ene["mni"]["wat"]["bay"]) * 100)",
)
println(
"CIFAR LeNet Wat, Accuracy: $(comp_exp["cif"]["wat"]["frq"]), Energy: $(comp_ene["cif"]["wat"]["frq"]), Ratio: $((comp_exp["cif"]["wat"]["frq"] / comp_ene["cif"]["wat"]["frq"]) * 100)",
)
println(
"CIFAR Bayes Wat, Accuracy: $(comp_exp["cif"]["wat"]["bay"]), Energy: $(comp_ene["cif"]["wat"]["bay"]), Ratio: $((comp_exp["cif"]["wat"]["bay"] / comp_ene["cif"]["wat"]["bay"]) * 100)",
)
println(
"MNIST LeNet Est, Accuracy: $(comp_exp["mni"]["est"]["frq"]), Energy: $(comp_ene["mni"]["est"]["frq"]), Ratio: $((comp_exp["mni"]["est"]["frq"] / comp_ene["mni"]["est"]["frq"]) * 100)",
)
println(
"MNIST Bayes Est, Accuracy: $(comp_exp["mni"]["est"]["bay"]), Energy: $(comp_ene["mni"]["est"]["bay"]), Ratio: $((comp_exp["mni"]["est"]["bay"] / comp_ene["mni"]["est"]["bay"]) * 100)",
)
println(
"CIFAR LeNet Est, Accuracy: $(comp_exp["cif"]["est"]["frq"]), Energy: $(comp_ene["cif"]["est"]["frq"]), Ratio: $((comp_exp["cif"]["est"]["frq"] / comp_ene["cif"]["est"]["frq"]) * 100)",
)
println(
"CIFAR Bayes Est, Accuracy: $(comp_exp["cif"]["est"]["bay"]), Energy: $(comp_ene["cif"]["est"]["bay"]), Ratio: $((comp_exp["cif"]["est"]["bay"] / comp_ene["cif"]["est"]["bay"]) * 100)",
)
#=
Final Metric
=#
mni_tot_frq_exp =
(
comp_exp["mni"]["100"]["frq"] +
comp_exp["mni"]["acc"]["frq"] +
comp_exp["mni"]["wat"]["frq"] +
comp_exp["mni"]["est"]["frq"]
) / 4
mni_tot_bay_exp =
(
comp_exp["mni"]["100"]["bay"] +
comp_exp["mni"]["acc"]["bay"] +
comp_exp["mni"]["wat"]["bay"] +
comp_exp["mni"]["est"]["bay"]
) / 4
cif_tot_frq_exp =
(
comp_exp["cif"]["100"]["frq"] +
comp_exp["cif"]["acc"]["frq"] +
comp_exp["cif"]["wat"]["frq"] +
comp_exp["cif"]["est"]["frq"]
) / 4
cif_tot_bay_exp =
(
comp_exp["cif"]["100"]["bay"] +
comp_exp["cif"]["acc"]["bay"] +
comp_exp["cif"]["wat"]["bay"] +
comp_exp["cif"]["est"]["bay"]
) / 4
mni_tot_frq_ene =
(
comp_ene["mni"]["100"]["frq"] +
comp_ene["mni"]["acc"]["frq"] +
comp_ene["mni"]["wat"]["frq"] +
comp_ene["mni"]["est"]["frq"]
) / 4
mni_tot_bay_ene =
(
comp_ene["mni"]["100"]["bay"] +
comp_ene["mni"]["acc"]["bay"] +
comp_ene["mni"]["wat"]["bay"] +
comp_ene["mni"]["est"]["bay"]
) / 4
cif_tot_frq_ene =
(
comp_ene["cif"]["100"]["frq"] +
comp_ene["cif"]["acc"]["frq"] +
comp_ene["cif"]["wat"]["frq"] +
comp_ene["cif"]["est"]["frq"]
) / 4
cif_tot_bay_ene =
(
comp_ene["cif"]["100"]["bay"] +
comp_ene["cif"]["acc"]["bay"] +
comp_ene["cif"]["wat"]["bay"] +
comp_ene["cif"]["est"]["bay"]
) / 4
println("LeNet-5 on MNIST Efficiency Metric: $((mni_tot_frq_exp * 100) / mni_tot_frq_ene)")
println("LeNet-5 on CIFAR Efficiency Metric: $((cif_tot_frq_exp * 100) / cif_tot_frq_ene)")
println("BCNN on MNIST Efficiency Metric: $((mni_tot_bay_exp * 100) / mni_tot_bay_ene)")
println("BCNN on CIFAR Efficiency Metric: $((cif_tot_bay_exp * 100) / cif_tot_bay_ene)")

508
energy_measure_dt.jl Executable file
View File

@ -0,0 +1,508 @@
include("aux_func.jl")
using PlotlyJS: savefig
using Statistics
using PlotlyJS
using Random
#=
Obtain energy data
=#
#mni_folder_1 = "ini_exp_data/"
mni_folder_100 = "exp_100_epochs/";
cif_folder_100 = "CIFAR_100_epoch/";
mni_folder_acc = "data_bounded/";
cif_folder_acc = "CIFAR_acc_bound/";
mni_folder_wat = "data_budget/";
cif_folder_wat = "CIFAR_energy_bound/";
mni_folder_est = "early_stop_res/";
cif_folder_est = "CIFAR_early_stop/";
bayes_model = "bayes";
freq_model = "freq";
w_type = "watt";
e_type = "exp";
#=
Load GPU data
=#
mni_100_bay_ene = getgpudata(mni_folder_100, bayes_model, w_type, "eo");
mni_100_fre_ene = getgpudata(mni_folder_100, freq_model, w_type, "eo");
cif_100_bay_ene = getgpudata(cif_folder_100, bayes_model, w_type, "eo");
cif_100_fre_ene = getgpudata(cif_folder_100, freq_model, w_type, "eo");
#=
Define CPU paths
=#
cif = "cifar";
mni = "mnist";
bay = "bayes";
wat = "cpu_watts";
frq = "freq";
ram = "ram_use";
_100 = "100";
acc = "acc";
es = "es";
wbud = "wbud";
#=
Load CPU data
=#
bay_cif_100_ene = readcpudata(cif, bay, wat, _100);
bay_mni_100_ene = readcpudata(mni, bay, wat, _100);
frq_cif_100_ene = readcpudata(cif, frq, wat, _100);
frq_mni_100_ene = readcpudata(mni, frq, wat, _100);
for s = 1:5
bay_cif_100_ene[s] = round.(getcpuwatt(bay_cif_100_ene[s]))
bay_mni_100_ene[s] = round.(getcpuwatt(bay_mni_100_ene[s]))
frq_cif_100_ene[s] = round.(getcpuwatt(frq_cif_100_ene[s]))
frq_mni_100_ene[s] = round.(getcpuwatt(frq_mni_100_ene[s]))
end
for s = 1:5
mni_100_bay_ene[s]["Ene"] = vcat(mni_100_bay_ene[s]["Ene"], bay_mni_100_ene[s])
mni_100_fre_ene[s]["Ene"] = vcat(mni_100_fre_ene[s]["Ene"], frq_mni_100_ene[s])
cif_100_bay_ene[s]["Ene"] = vcat(cif_100_bay_ene[s]["Ene"], bay_cif_100_ene[s])
cif_100_fre_ene[s]["Ene"] = vcat(cif_100_fre_ene[s]["Ene"], frq_cif_100_ene[s])
end
ene_data = Dict("mni" => Dict("bay" => Dict(1 => [],2 => [],3 => [],4 => [],5 => []),
"fre" => Dict(1 => [],2 => [],3 => [],4 => [],5 => [])),
"cif" => Dict("bay" => Dict(1 => [],2 => [],3 => [],4 => [],5 => []),
"fre" => Dict(1 => [],2 => [],3 => [],4 => [],5 => [])))
#ene_data = Dict("mni" => Dict("bay" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0),
# "fre" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0)),
# "cif" => Dict("bay" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0),
# "fre" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0)))
for s = 1:5
ene_data["mni"]["bay"][s] = sum.(join_vectors(resize1(makechunks(shuffle(mni_100_bay_ene[s]["Ene"]), 100))))
ene_data["mni"]["fre"][s] = sum.(join_vectors(resize1(makechunks(shuffle(mni_100_fre_ene[s]["Ene"]), 100))))
ene_data["cif"]["bay"][s] = sum.(join_vectors(resize1(makechunks(shuffle(cif_100_bay_ene[s]["Ene"]), 100))))
ene_data["cif"]["fre"][s] = sum.(join_vectors(resize1(makechunks(shuffle(cif_100_fre_ene[s]["Ene"]), 100))))
end
#for s = 1:5
# ene_data["mni"]["bay"][s] = sum(mni_100_bay_ene[s]["Ene"]) / 100
# ene_data["mni"]["fre"][s] = sum(mni_100_fre_ene[s]["Ene"]) / 100
# ene_data["cif"]["bay"][s] = sum(cif_100_bay_ene[s]["Ene"]) / 100
# ene_data["cif"]["fre"][s] = sum(cif_100_fre_ene[s]["Ene"]) / 100
#end
#=
Load Accuracy data
=#
mni_100_bay_exp = getgpudata(mni_folder_100, bayes_model, e_type, "eo");
mni_100_fre_exp = getgpudata(mni_folder_100, freq_model, e_type, "eo");
cif_100_bay_exp = getgpudata(cif_folder_100, bayes_model, e_type, "eo");
cif_100_fre_exp = getgpudata(cif_folder_100, freq_model, e_type, "eo");
for i = 1:5
mni_100_bay_exp[i] = expdatatodict(mni_100_bay_exp[i])
mni_100_fre_exp[i] = expdatatodict(mni_100_fre_exp[i])
cif_100_bay_exp[i] = expdatatodict(cif_100_bay_exp[i])
cif_100_fre_exp[i] = expdatatodict(cif_100_fre_exp[i])
end
original_vectors = Dict("LeNet" => Dict("MNIST" => Dict(1 => [],2 => [],3 => [],4 => [],5 => []),
"CIFAR" => Dict(1 => [],2 => [],3 => [],4 => [],5 => [])),
"BCNN" => Dict("MNIST" => Dict(1 => [],2 => [],3 => [],4 => [],5 => []),
"CIFAR" => Dict(1 => [],2 => [],3 => [],4 => [],5 => [])))
#=
delta_vectors = Dict("LeNet" => Dict("MNIST" => Dict(1 => [],2 => [],3 => [],4 => [],5 => []),
"CIFAR" => Dict(1 => [],2 => [],3 => [],4 => [],5 => [])),
"BCNN" => Dict("MNIST" => Dict(1 => [],2 => [],3 => [],4 => [],5 => []),
"CIFAR" => Dict(1 => [],2 => [],3 => [],4 => [],5 => [])))
delta2_vectors = Dict("LeNet" => Dict("MNIST" => Dict(1 => [],2 => [],3 => [],4 => [],5 => []),
"CIFAR" => Dict(1 => [],2 => [],3 => [],4 => [],5 => [])),
"BCNN" => Dict("MNIST" => Dict(1 => [],2 => [],3 => [],4 => [],5 => []),
"CIFAR" => Dict(1 => [],2 => [],3 => [],4 => [],5 => [])))
=#
delta_values = Dict("LeNet" => Dict("MNIST" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0),
"CIFAR" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0)),
"BCNN" => Dict("MNIST" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0),
"CIFAR" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0)))
for i in 1:5
original_vectors["LeNet"]["MNIST"][i] = (mni_100_fre_exp[i]["acc"] ./ ene_data["mni"]["fre"][i]) * 100
original_vectors["BCNN"]["MNIST"][i] = (mni_100_bay_exp[i]["acc"] ./ ene_data["mni"]["bay"][i]) * 100
original_vectors["LeNet"]["CIFAR"][i] = (cif_100_fre_exp[i]["acc"] ./ ene_data["cif"]["fre"][i]) * 100
original_vectors["BCNN"]["CIFAR"][i] = (cif_100_bay_exp[i]["acc"] ./ ene_data["cif"]["bay"][i]) * 100
end
for i in 1:5
delta_values["LeNet"]["MNIST"][i] = mean(approx_δ(original_vectors["LeNet"]["MNIST"][i]))
delta_values["BCNN"]["MNIST"][i] = mean(approx_δ(original_vectors["BCNN"]["MNIST"][i]))
delta_values["LeNet"]["CIFAR"][i] = mean(approx_δ(original_vectors["LeNet"]["CIFAR"][i]))
delta_values["BCNN"]["CIFAR"][i] = mean(approx_δ(original_vectors["BCNN"]["CIFAR"][i]))
end
lenet_mnist = zeros(5)
bcnn_mnist = zeros(5)
lenet_cifar = zeros(5)
bcnn_cifar = zeros(5)
#=
for i in 1:5
println("LeNet MNIST Size $(i): $(delta_values["LeNet"]["MNIST"][i])")
println("BCNN MNIST Size $(i): $(delta_values["BCNN"]["MNIST"][i])")
println("LeNet CIFAR Size $(i): $(delta_values["LeNet"]["CIFAR"][i])")
println("BCNN CIFAR Size $(i): $(delta_values["BCNN"]["CIFAR"][i])")
end
=#
for i in 1:5
lenet_mnist[i] = delta_values["LeNet"]["MNIST"][i]
bcnn_mnist[i] = delta_values["BCNN"]["MNIST"][i]
lenet_cifar[i] = delta_values["LeNet"]["CIFAR"][i]
bcnn_cifar[i] = delta_values["BCNN"]["CIFAR"][i]
end
println("MNIST LeNet Δ: $(mean(lenet_mnist))")
println("MNIST BCNN Δ: $(mean(bcnn_mnist))")
println("CIFAR LeNet Δ: $(mean(lenet_cifar))")
println("CIFAR BCNN Δ: $(mean(bcnn_cifar))")
#=
en_plot = plot(
[
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta_vectors["LeNet"]["MNIST"][1],
name = "LeNet 1 MNIST",
marker = attr(symbol = 4, color = "rgb(211,120,000)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta_vectors["LeNet"]["MNIST"][2],
name = "LeNet 2 MNIST",
marker = attr(symbol = 17, color = "rgb(255,170,017)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta_vectors["LeNet"]["MNIST"][3],
name = "LeNet 3 MNIST",
marker = attr(symbol = 2, color = "rgb(255,187,034)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta_vectors["LeNet"]["MNIST"][4],
name = "LeNet 4 MNIST",
marker = attr(symbol = 0, color = "rgb(255,204,051)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta_vectors["LeNet"]["MNIST"][5],
name = "LeNet 5 MNIST",
marker = attr(symbol = 4, color = "rgb(255,195,001)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta_vectors["LeNet"]["CIFAR"][1],
name = "LeNet 1 CIFAR",
marker = attr(symbol = 4, color = "rgb(055,033,240)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta_vectors["LeNet"]["CIFAR"][2],
name = "LeNet 2 CIFAR",
marker = attr(symbol = 4, color = "rgb(018,094,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta_vectors["LeNet"]["CIFAR"][3],
name = "LeNet 3 CIFAR",
marker = attr(symbol = 4, color = "rgb(018,126,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta_vectors["LeNet"]["CIFAR"][4],
name = "LeNet 4 CIFAR",
marker = attr(symbol = 4, color = "rgb(018,150,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta_vectors["LeNet"]["CIFAR"][5],
name = "LeNet 5 CIFAR",
marker = attr(symbol = 4, color = "rgb(018,167,219)", line_width = 1.0),
),
],
Layout(
mode = "overlay",
xaxis_tickangle = -45,
yaxis_title_text = "ΔEfficency",
xaxis_title_text = "Epoch";
opacity = 0.9,
xaxis_range = [0, 20],
#xaxis_range = [-1, 5],
#xaxis_type = "category",
),
)
savefig(en_plot, "lenet_dE_epoch.png")
en_plot = plot(
[
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta_vectors["BCNN"]["MNIST"][1],
name = "BCNN 1 MNIST",
marker = attr(symbol = 4, color = "rgb(211,120,000)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta_vectors["BCNN"]["MNIST"][2],
name = "BCNN 2 MNIST",
marker = attr(symbol = 17, color = "rgb(255,170,017)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta_vectors["BCNN"]["MNIST"][3],
name = "BCNN 3 MNIST",
marker = attr(symbol = 2, color = "rgb(255,187,034)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta_vectors["BCNN"]["MNIST"][4],
name = "BCNN 4 MNIST",
marker = attr(symbol = 0, color = "rgb(255,204,051)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta_vectors["BCNN"]["MNIST"][5],
name = "BCNN 5 MNIST",
marker = attr(symbol = 4, color = "rgb(255,195,001)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta_vectors["BCNN"]["CIFAR"][1],
name = "BCNN 1 CIFAR",
marker = attr(symbol = 4, color = "rgb(055,033,240)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta_vectors["BCNN"]["CIFAR"][2],
name = "BCNN 2 CIFAR",
marker = attr(symbol = 4, color = "rgb(018,094,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta_vectors["BCNN"]["CIFAR"][3],
name = "BCNN 3 CIFAR",
marker = attr(symbol = 4, color = "rgb(018,126,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta_vectors["BCNN"]["CIFAR"][4],
name = "BCNN 4 CIFAR",
marker = attr(symbol = 4, color = "rgb(018,150,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta_vectors["BCNN"]["CIFAR"][5],
name = "BCNN 5 CIFAR",
marker = attr(symbol = 4, color = "rgb(018,167,219)", line_width = 1.0),
),
],
Layout(
mode = "overlay",
xaxis_tickangle = -45,
yaxis_title_text = "ΔEfficency",
xaxis_title_text = "Epoch";
opacity = 0.9,
xaxis_range = [0, 20],
#xaxis_range = [-1, 5],
#xaxis_type = "category",
),
)
savefig(en_plot, "bcnn_dE_epoch.png")
en_plot = plot(
[
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta2_vectors["LeNet"]["MNIST"][1],
name = "LeNet 1 MNIST",
marker = attr(symbol = 4, color = "rgb(211,120,000)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta2_vectors["LeNet"]["MNIST"][2],
name = "LeNet 2 MNIST",
marker = attr(symbol = 17, color = "rgb(255,170,017)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta2_vectors["LeNet"]["MNIST"][3],
name = "LeNet 3 MNIST",
marker = attr(symbol = 2, color = "rgb(255,187,034)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta2_vectors["LeNet"]["MNIST"][4],
name = "LeNet 4 MNIST",
marker = attr(symbol = 0, color = "rgb(255,204,051)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta2_vectors["LeNet"]["MNIST"][5],
name = "LeNet 5 MNIST",
marker = attr(symbol = 4, color = "rgb(255,195,001)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta2_vectors["LeNet"]["CIFAR"][1],
name = "LeNet 1 CIFAR",
marker = attr(symbol = 4, color = "rgb(055,033,240)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta2_vectors["LeNet"]["CIFAR"][2],
name = "LeNet 2 CIFAR",
marker = attr(symbol = 4, color = "rgb(018,094,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta2_vectors["LeNet"]["CIFAR"][3],
name = "LeNet 3 CIFAR",
marker = attr(symbol = 4, color = "rgb(018,126,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta2_vectors["LeNet"]["CIFAR"][4],
name = "LeNet 4 CIFAR",
marker = attr(symbol = 4, color = "rgb(018,150,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta2_vectors["LeNet"]["CIFAR"][5],
name = "LeNet 5 CIFAR",
marker = attr(symbol = 4, color = "rgb(018,167,219)", line_width = 1.0),
),
],
Layout(
mode = "overlay",
xaxis_tickangle = -45,
yaxis_title_text = "ΔΔEfficency",
xaxis_title_text = "Epoch";
opacity = 0.9,
xaxis_range = [0, 20],
#xaxis_range = [-1, 5],
#xaxis_type = "category",
),
)
savefig(en_plot, "lenet_ddE_epoch.png")
en_plot = plot(
[
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta2_vectors["BCNN"]["MNIST"][1],
name = "BCNN 1 MNIST",
marker = attr(symbol = 4, color = "rgb(211,120,000)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta2_vectors["BCNN"]["MNIST"][2],
name = "BCNN 2 MNIST",
marker = attr(symbol = 17, color = "rgb(255,170,017)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta2_vectors["BCNN"]["MNIST"][3],
name = "BCNN 3 MNIST",
marker = attr(symbol = 2, color = "rgb(255,187,034)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta2_vectors["BCNN"]["MNIST"][4],
name = "BCNN 4 MNIST",
marker = attr(symbol = 0, color = "rgb(255,204,051)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta2_vectors["BCNN"]["MNIST"][5],
name = "BCNN 5 MNIST",
marker = attr(symbol = 4, color = "rgb(255,195,001)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta2_vectors["BCNN"]["CIFAR"][1],
name = "BCNN 1 CIFAR",
marker = attr(symbol = 4, color = "rgb(055,033,240)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta2_vectors["BCNN"]["CIFAR"][2],
name = "BCNN 2 CIFAR",
marker = attr(symbol = 4, color = "rgb(018,094,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta2_vectors["BCNN"]["CIFAR"][3],
name = "BCNN 3 CIFAR",
marker = attr(symbol = 4, color = "rgb(018,126,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta2_vectors["BCNN"]["CIFAR"][4],
name = "BCNN 4 CIFAR",
marker = attr(symbol = 4, color = "rgb(018,150,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = delta2_vectors["BCNN"]["CIFAR"][5],
name = "BCNN 5 CIFAR",
marker = attr(symbol = 4, color = "rgb(018,167,219)", line_width = 1.0),
),
],
Layout(
mode = "overlay",
xaxis_tickangle = -45,
yaxis_title_text = "ΔΔEfficency",
xaxis_title_text = "Epoch";
opacity = 0.9,
xaxis_range = [0, 20],
#xaxis_range = [-1, 5],
#xaxis_type = "category",
),
)
savefig(en_plot, "bcnn_ddE_epoch.png")
=#

1052
energy_measure_sum.jl Executable file
View File

File diff suppressed because it is too large Load Diff

399
energy_measure_sum_df.jl Normal file
View File

@ -0,0 +1,399 @@
include("aux_func.jl")
using LinearAlgebra
using Statistics
using DataFrames
using Latexify
#=
Obtain energy data
=#
#mni_folder_1 = "ini_exp_data/"
mni_folder_100 = "exp_100_epochs/";
cif_folder_100 = "CIFAR_100_epoch/";
mni_folder_acc = "data_bounded/";
cif_folder_acc = "CIFAR_acc_bound/";
mni_folder_wat = "data_budget/";
cif_folder_wat = "CIFAR_energy_bound/";
mni_folder_est = "early_stop_res/";
cif_folder_est = "CIFAR_early_stop/";
bayes_model = "bayes";
freq_model = "freq";
w_type = "watt";
e_type = "exp";
#=
Load GPU data
=#
mni_100_bay_ene = getgpudata(mni_folder_100, bayes_model, w_type, "eo");
mni_100_fre_ene = getgpudata(mni_folder_100, freq_model, w_type, "eo");
cif_100_bay_ene = getgpudata(cif_folder_100, bayes_model, w_type, "eo");
cif_100_fre_ene = getgpudata(cif_folder_100, freq_model, w_type, "eo");
mni_acc_bay_ene = getgpudata(mni_folder_acc, bayes_model, w_type, "eo");
mni_acc_fre_ene = getgpudata(mni_folder_acc, freq_model, w_type, "eo");
cif_acc_bay_ene = getgpudata(cif_folder_acc, bayes_model, w_type, "eo");
cif_acc_fre_ene = getgpudata(cif_folder_acc, freq_model, w_type, "eo");
mni_wat_bay_ene = getgpudata(mni_folder_wat, bayes_model, w_type, "eo");
mni_wat_fre_ene = getgpudata(mni_folder_wat, freq_model, w_type, "eo");
cif_wat_bay_ene = getgpudata(cif_folder_wat, bayes_model, w_type, "eo");
cif_wat_fre_ene = getgpudata(cif_folder_wat, freq_model, w_type, "eo");
mni_est_bay_ene = getgpudata(mni_folder_est, bayes_model, w_type, "eo");
mni_est_fre_ene = getgpudata(mni_folder_est, freq_model, w_type, "eo");
cif_est_bay_ene = getgpudata(cif_folder_est, bayes_model, w_type, "eo");
cif_est_fre_ene = getgpudata(cif_folder_est, freq_model, w_type, "eo");
#=
Define CPU paths
=#
cif = "cifar";
mni = "mnist";
bay = "bayes";
wat = "cpu_watts";
frq = "freq";
ram = "ram_use";
_100 = "100";
acc = "acc";
es = "es";
wbud = "wbud";
#=
Load CPU data
=#
bay_cif_100_ene = readcpudata(cif, bay, wat, _100);
bay_mni_100_ene = readcpudata(mni, bay, wat, _100);
frq_cif_100_ene = readcpudata(cif, frq, wat, _100);
frq_mni_100_ene = readcpudata(mni, frq, wat, _100);
bay_cif_acc_ene = readcpudata(cif, bay, wat, acc);
bay_mni_acc_ene = readcpudata(mni, bay, wat, acc);
frq_cif_acc_ene = readcpudata(cif, frq, wat, acc);
frq_mni_acc_ene = readcpudata(mni, frq, wat, acc);
bay_cif_wbu_ene = readcpudata(cif, bay, wat, wbud);
bay_mni_wbu_ene = readcpudata(mni, bay, wat, wbud);
frq_cif_wbu_ene = readcpudata(cif, frq, wat, wbud);
frq_mni_wbu_ene = readcpudata(mni, frq, wat, wbud);
bay_mni_est_ene = readcpudata(mni, bay, wat, es);
bay_cif_est_ene = readcpudata(cif, bay, wat, es);
frq_cif_est_ene = readcpudata(cif, frq, wat, es);
frq_mni_est_ene = readcpudata(mni, frq, wat, es);
for s = 1:5
bay_cif_acc_ene[s] = round.(getcpuwatt(bay_cif_acc_ene[s]))
bay_cif_est_ene[s] = round.(getcpuwatt(bay_cif_est_ene[s]))
bay_cif_wbu_ene[s] = round.(getcpuwatt(bay_cif_wbu_ene[s]))
bay_cif_100_ene[s] = round.(getcpuwatt(bay_cif_100_ene[s]))
bay_mni_acc_ene[s] = round.(getcpuwatt(bay_mni_acc_ene[s]))
bay_mni_est_ene[s] = round.(getcpuwatt(bay_mni_est_ene[s]))
bay_mni_wbu_ene[s] = round.(getcpuwatt(bay_mni_wbu_ene[s]))
bay_mni_100_ene[s] = round.(getcpuwatt(bay_mni_100_ene[s]))
frq_cif_acc_ene[s] = round.(getcpuwatt(frq_cif_acc_ene[s]))
frq_cif_est_ene[s] = round.(getcpuwatt(frq_cif_est_ene[s]))
frq_cif_wbu_ene[s] = round.(getcpuwatt(frq_cif_wbu_ene[s]))
frq_cif_100_ene[s] = round.(getcpuwatt(frq_cif_100_ene[s]))
frq_mni_acc_ene[s] = round.(getcpuwatt(frq_mni_acc_ene[s]))
frq_mni_est_ene[s] = round.(getcpuwatt(frq_mni_est_ene[s]))
frq_mni_wbu_ene[s] = round.(getcpuwatt(frq_mni_wbu_ene[s]))
frq_mni_100_ene[s] = round.(getcpuwatt(frq_mni_100_ene[s]))
end
for s = 1:5
mni_100_bay_ene[s]["Ene"] = vcat(mni_100_bay_ene[s]["Ene"], bay_mni_100_ene[s])
mni_100_fre_ene[s]["Ene"] = vcat(mni_100_fre_ene[s]["Ene"], frq_mni_100_ene[s])
cif_100_bay_ene[s]["Ene"] = vcat(cif_100_bay_ene[s]["Ene"], bay_cif_100_ene[s])
cif_100_fre_ene[s]["Ene"] = vcat(cif_100_fre_ene[s]["Ene"], frq_cif_100_ene[s])
mni_acc_bay_ene[s]["Ene"] = vcat(mni_acc_bay_ene[s]["Ene"], bay_mni_acc_ene[s])
mni_acc_fre_ene[s]["Ene"] = vcat(mni_acc_fre_ene[s]["Ene"], frq_mni_acc_ene[s])
cif_acc_bay_ene[s]["Ene"] = vcat(cif_acc_bay_ene[s]["Ene"], bay_cif_acc_ene[s])
cif_acc_fre_ene[s]["Ene"] = vcat(cif_acc_fre_ene[s]["Ene"], frq_cif_acc_ene[s])
mni_wat_bay_ene[s]["Ene"] = vcat(mni_wat_bay_ene[s]["Ene"], bay_mni_wbu_ene[s])
mni_wat_fre_ene[s]["Ene"] = vcat(mni_wat_fre_ene[s]["Ene"], frq_mni_wbu_ene[s])
cif_wat_bay_ene[s]["Ene"] = vcat(cif_wat_bay_ene[s]["Ene"], bay_cif_wbu_ene[s])
cif_wat_fre_ene[s]["Ene"] = vcat(cif_wat_fre_ene[s]["Ene"], frq_cif_wbu_ene[s])
mni_est_bay_ene[s]["Ene"] = vcat(mni_est_bay_ene[s]["Ene"], bay_mni_est_ene[s])
mni_est_fre_ene[s]["Ene"] = vcat(mni_est_fre_ene[s]["Ene"], frq_mni_est_ene[s])
cif_est_bay_ene[s]["Ene"] = vcat(cif_est_bay_ene[s]["Ene"], bay_cif_est_ene[s])
cif_est_fre_ene[s]["Ene"] = vcat(cif_est_fre_ene[s]["Ene"], frq_cif_est_ene[s])
end
mnist_freq_100_ene = zeros(5)
mnist_freq_est_ene = zeros(5)
mnist_freq_wat_ene = zeros(5)
mnist_freq_acc_ene = zeros(5)
mnist_baye_100_ene = zeros(5)
mnist_baye_est_ene = zeros(5)
mnist_baye_wat_ene = zeros(5)
mnist_baye_acc_ene = zeros(5)
cifar_freq_100_ene = zeros(5)
cifar_freq_est_ene = zeros(5)
cifar_freq_wat_ene = zeros(5)
cifar_freq_acc_ene = zeros(5)
cifar_baye_100_ene = zeros(5)
cifar_baye_est_ene = zeros(5)
cifar_baye_wat_ene = zeros(5)
cifar_baye_acc_ene = zeros(5)
for s = 1:5
mnist_baye_100_ene[s] = sum(mni_100_bay_ene[s]["Ene"])
mnist_freq_100_ene[s] = sum(mni_100_fre_ene[s]["Ene"])
cifar_baye_100_ene[s] = sum(cif_100_bay_ene[s]["Ene"])
cifar_freq_100_ene[s] = sum(cif_100_fre_ene[s]["Ene"])
mnist_baye_acc_ene[s] = sum(mni_acc_bay_ene[s]["Ene"])
mnist_freq_acc_ene[s] = sum(mni_acc_fre_ene[s]["Ene"])
cifar_baye_acc_ene[s] = sum(cif_acc_bay_ene[s]["Ene"])
cifar_freq_acc_ene[s] = sum(cif_acc_fre_ene[s]["Ene"])
mnist_baye_wat_ene[s] = sum(mni_wat_bay_ene[s]["Ene"])
mnist_freq_wat_ene[s] = sum(mni_wat_fre_ene[s]["Ene"])
cifar_baye_wat_ene[s] = sum(cif_wat_bay_ene[s]["Ene"])
cifar_freq_wat_ene[s] = sum(cif_wat_fre_ene[s]["Ene"])
mnist_baye_est_ene[s] = sum(mni_est_bay_ene[s]["Ene"])
mnist_freq_est_ene[s] = sum(mni_est_fre_ene[s]["Ene"])
cifar_baye_est_ene[s] = sum(cif_est_bay_ene[s]["Ene"])
cifar_freq_est_ene[s] = sum(cif_est_fre_ene[s]["Ene"])
end
#=
Load Accuracy data
=#
mni_100_bay_exp = getgpudata(mni_folder_100, bayes_model, e_type, "eo");
mni_100_fre_exp = getgpudata(mni_folder_100, freq_model, e_type, "eo");
cif_100_bay_exp = getgpudata(cif_folder_100, bayes_model, e_type, "eo");
cif_100_fre_exp = getgpudata(cif_folder_100, freq_model, e_type, "eo");
mni_acc_bay_exp = getgpudata(mni_folder_acc, bayes_model, e_type, "eo");
mni_acc_fre_exp = getgpudata(mni_folder_acc, freq_model, e_type, "eo");
cif_acc_bay_exp = getgpudata(cif_folder_acc, bayes_model, e_type, "eo");
cif_acc_fre_exp = getgpudata(cif_folder_acc, freq_model, e_type, "eo");
mni_wat_bay_exp = getgpudata(mni_folder_wat, bayes_model, e_type, "eo");
mni_wat_fre_exp = getgpudata(mni_folder_wat, freq_model, e_type, "eo");
cif_wat_bay_exp = getgpudata(cif_folder_wat, bayes_model, e_type, "eo");
cif_wat_fre_exp = getgpudata(cif_folder_wat, freq_model, e_type, "eo");
mni_est_bay_exp = getgpudata(mni_folder_est, bayes_model, e_type, "eo");
mni_est_fre_exp = getgpudata(mni_folder_est, freq_model, e_type, "eo");
cif_est_bay_exp = getgpudata(cif_folder_est, bayes_model, e_type, "eo");
cif_est_fre_exp = getgpudata(cif_folder_est, freq_model, e_type, "eo");
for i = 1:5
mni_100_bay_exp[i] = expdatatodict(mni_100_bay_exp[i])
mni_100_fre_exp[i] = expdatatodict(mni_100_fre_exp[i])
cif_100_bay_exp[i] = expdatatodict(cif_100_bay_exp[i])
cif_100_fre_exp[i] = expdatatodict(cif_100_fre_exp[i])
mni_acc_bay_exp[i] = expdatatodict(mni_acc_bay_exp[i])
mni_acc_fre_exp[i] = expdatatodict(mni_acc_fre_exp[i])
cif_acc_bay_exp[i] = expdatatodict(cif_acc_bay_exp[i])
cif_acc_fre_exp[i] = expdatatodict(cif_acc_fre_exp[i])
mni_wat_bay_exp[i] = expdatatodict(mni_wat_bay_exp[i])
mni_wat_fre_exp[i] = expdatatodict(mni_wat_fre_exp[i])
cif_wat_bay_exp[i] = expdatatodict(cif_wat_bay_exp[i])
cif_wat_fre_exp[i] = expdatatodict(cif_wat_fre_exp[i])
mni_est_bay_exp[i] = expdatatodict(mni_est_bay_exp[i])
mni_est_fre_exp[i] = expdatatodict(mni_est_fre_exp[i])
cif_est_bay_exp[i] = expdatatodict(cif_est_bay_exp[i])
cif_est_fre_exp[i] = expdatatodict(cif_est_fre_exp[i])
end
mnist_freq_100_exp = zeros(5)
mnist_freq_est_exp = zeros(5)
mnist_freq_wat_exp = zeros(5)
mnist_freq_acc_exp = zeros(5)
mnist_baye_100_exp = zeros(5)
mnist_baye_est_exp = zeros(5)
mnist_baye_wat_exp = zeros(5)
mnist_baye_acc_exp = zeros(5)
cifar_freq_100_exp = zeros(5)
cifar_freq_est_exp = zeros(5)
cifar_freq_wat_exp = zeros(5)
cifar_freq_acc_exp = zeros(5)
cifar_baye_100_exp = zeros(5)
cifar_baye_est_exp = zeros(5)
cifar_baye_wat_exp = zeros(5)
cifar_baye_acc_exp = zeros(5)
for s = 1:5
mnist_baye_100_exp[s] = (mean(mni_100_bay_exp[s]["acc"]) + mean(mni_100_bay_exp[s]["pre"])) / 2
mnist_freq_100_exp[s] = (mean(mni_100_fre_exp[s]["acc"]) + mean(mni_100_fre_exp[s]["pre"])) / 2
cifar_baye_100_exp[s] = (mean(cif_100_bay_exp[s]["acc"]) + mean(cif_100_bay_exp[s]["pre"])) / 2
cifar_freq_100_exp[s] = (mean(cif_100_fre_exp[s]["acc"]) + mean(cif_100_fre_exp[s]["pre"])) / 2
mnist_baye_acc_exp[s] = (mean(mni_acc_bay_exp[s]["acc"]) + mean(mni_acc_bay_exp[s]["pre"])) / 2
mnist_freq_acc_exp[s] = (mean(mni_acc_fre_exp[s]["acc"]) + mean(mni_acc_fre_exp[s]["pre"])) / 2
cifar_baye_acc_exp[s] = (mean(cif_acc_bay_exp[s]["acc"]) + mean(cif_acc_bay_exp[s]["pre"])) / 2
cifar_freq_acc_exp[s] = (mean(cif_acc_fre_exp[s]["acc"]) + mean(cif_acc_fre_exp[s]["pre"])) / 2
mnist_baye_wat_exp[s] = (mean(mni_wat_bay_exp[s]["acc"]) + mean(mni_wat_bay_exp[s]["pre"])) / 2
mnist_freq_wat_exp[s] = (mean(mni_wat_fre_exp[s]["acc"]) + mean(mni_wat_fre_exp[s]["pre"])) / 2
cifar_baye_wat_exp[s] = (mean(cif_wat_bay_exp[s]["acc"]) + mean(cif_wat_bay_exp[s]["pre"])) / 2
cifar_freq_wat_exp[s] = (mean(cif_wat_fre_exp[s]["acc"]) + mean(cif_wat_fre_exp[s]["pre"])) / 2
mnist_baye_est_exp[s] = (mean(mni_est_bay_exp[s]["acc"]) + mean(mni_est_bay_exp[s]["pre"])) / 2
mnist_freq_est_exp[s] = (mean(mni_est_fre_exp[s]["acc"]) + mean(mni_est_fre_exp[s]["pre"])) / 2
cifar_baye_est_exp[s] = (mean(cif_est_bay_exp[s]["acc"]) + mean(cif_est_bay_exp[s]["pre"])) / 2
cifar_freq_est_exp[s] = (mean(cif_est_fre_exp[s]["acc"]) + mean(cif_est_fre_exp[s]["pre"])) / 2
end
all_data_exp = Dict(
"mni" => Dict(
1 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
2 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
3 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
4 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
5 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
),
"cif" => Dict(
1 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
2 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
3 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
4 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
5 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
),
)
mnist_freq_100_df = DataFrame(Accuracy=mnist_freq_100_exp,Energy=mnist_freq_100_ene)
mnist_freq_est_df = DataFrame(Accuracy=mnist_freq_est_exp,Energy=mnist_freq_est_ene)
mnist_freq_wat_df = DataFrame(Accuracy=mnist_freq_wat_exp,Energy=mnist_freq_wat_ene)
mnist_freq_acc_df = DataFrame(Accuracy=mnist_freq_acc_exp,Energy=mnist_freq_acc_ene)
mnist_baye_100_df = DataFrame(Accuracy=mnist_baye_100_exp,Energy=mnist_baye_100_ene)
mnist_baye_est_df = DataFrame(Accuracy=mnist_baye_est_exp,Energy=mnist_baye_est_ene)
mnist_baye_wat_df = DataFrame(Accuracy=mnist_baye_wat_exp,Energy=mnist_baye_wat_ene)
mnist_baye_acc_df = DataFrame(Accuracy=mnist_baye_acc_exp,Energy=mnist_baye_acc_ene)
cifar_freq_100_df = DataFrame(Accuracy=cifar_freq_100_exp,Energy=cifar_freq_100_ene)
cifar_freq_est_df = DataFrame(Accuracy=cifar_freq_est_exp,Energy=cifar_freq_est_ene)
cifar_freq_wat_df = DataFrame(Accuracy=cifar_freq_wat_exp,Energy=cifar_freq_wat_ene)
cifar_freq_acc_df = DataFrame(Accuracy=cifar_freq_acc_exp,Energy=cifar_freq_acc_ene)
cifar_baye_100_df = DataFrame(Accuracy=cifar_baye_100_exp,Energy=cifar_baye_100_ene)
cifar_baye_est_df = DataFrame(Accuracy=cifar_baye_est_exp,Energy=cifar_baye_est_ene)
cifar_baye_wat_df = DataFrame(Accuracy=cifar_baye_wat_exp,Energy=cifar_baye_wat_ene)
cifar_baye_acc_df = DataFrame(Accuracy=cifar_baye_acc_exp,Energy=cifar_baye_acc_ene)
println("MNIST 100 Stopping")
latexify(mnist_baye_100_df,env=:table) |> print
latexify(mnist_freq_100_df,env=:table) |> print
println("CIFAR 100 Stopping")
latexify(cifar_baye_100_df,env=:table) |> print
latexify(cifar_freq_100_df,env=:table) |> print
println("MNIST Early Stopping")
latexify(mnist_baye_est_df,env=:table) |> print
latexify(mnist_freq_est_df,env=:table) |> print
println("CIFAR Early Stopping")
latexify(cifar_baye_est_df,env=:table) |> print
latexify(cifar_freq_est_df,env=:table) |> print
println("MNIST Energy bound")
latexify(mnist_baye_wat_df,env=:table) |> print
latexify(mnist_freq_wat_df,env=:table) |> print
println("CIFAR Energy bound")
latexify(cifar_baye_wat_df,env=:table) |> print
latexify(cifar_freq_wat_df,env=:table) |> print
println("MNIST Accuracy bound")
latexify(mnist_baye_acc_df,env=:table) |> print
latexify(mnist_freq_acc_df,env=:table) |> print
println("CIFAR Accuracy bound")
latexify(cifar_baye_acc_df,env=:table) |> print
latexify(cifar_freq_acc_df,env=:table) |> print

367
flop_scale.jl Executable file
View File

@ -0,0 +1,367 @@
###########################################################################
# #
# FLOPS Computations #
# #
###########################################################################
using PlotlyJS
using PlotlyJS: savefig
flops_bcnn_100_mnist =
[5.5e11 * 100, 5.5e11 * 200, 5.5e11 * 300, 5.5e11 * 400, 5.5e11 * 500]
flops_freq_100_mnist =
[2.58e11 * 100, 2.58e11 * 200, 2.58e11 * 300, 2.58e11 * 400, 2.58e11 * 500]
flops_bcnn_est_mnist =
[5.5e11 * 65, 5.5e11 * 21 * 2, 5.5e11 * 37 * 3, 5.5e11 * 53 * 4, 5.5e11 * 65 * 5]
flops_freq_est_mnist =
[2.58e11 * 16, 2.58e11 * 12 * 2, 2.58e11 * 56 * 3, 2.58e11 * 28 * 4, 2.58e11 * 20 * 5]
flops_bcnn_wat_mnist =
[5.5e11 * 19, 5.5e11 * 2 * 13, 5.5e11 * 3 * 10, 5.5e11 * 4 * 8, 5.5e11 * 5 * 6]
flops_freq_wat_mnist =
[2.58e11 * 43, 2.58e11 * 2 * 39, 2.58e11 * 3 * 27, 2.58e11 * 4 * 23, 2.58e11 * 5 * 21]
flops_bcnn_acc_mnist =
[5.5e11 * 72, 5.5e11 * 2 * 69, 5.5e11 * 3 * 69, 5.5e11 * 4 * 77, 5.5e11 * 5 * 80]
flops_freq_acc_mnist =
[2.58e11 * 12, 2.58e11 * 2 * 8, 2.58e11 * 3 * 6, 2.58e11 * 4 * 6, 2.58e11 * 5 * 4]
flops_bcnn_100_cifar = [
5.5e11 * 100 * 16.3,
5.5e11 * 200 * 16.3,
5.5e11 * 300 * 16.3,
5.5e11 * 400 * 16.3,
5.5e11 * 500 * 16.3,
]
flops_freq_100_cifar = [
2.58e11 * 100 * 16.3,
2.58e11 * 200 * 16.3,
2.58e11 * 300 * 16.3,
2.58e11 * 400 * 16.3,
2.58e11 * 500 * 16.3,
]
flops_bcnn_est_cifar = [
5.5e11 * 61 * 16.3,
5.5e11 * 2 * 41 * 16.3,
5.5e11 * 3 * 41 * 16.3,
5.5e11 * 4 * 21 * 16.3,
5.5e11 * 5 * 81 * 16.3,
]
flops_freq_est_cifar = [
2.58e11 * 56 * 16.3,
2.58e11 * 2 * 40 * 16.3,
2.58e11 * 3 * 24 * 16.3,
2.58e11 * 4 * 24 * 16.3,
2.58e11 * 5 * 28 * 16.3,
]
flops_bcnn_wat_cifar = [
5.5e11 * 16.3 * 19,
5.5e11 * 2 * 16.3 * 14,
5.5e11 * 3 * 16.3 * 11,
5.5e11 * 4 * 16.3 * 9,
5.5e11 * 5 * 16.3 * 7,
]
flops_freq_wat_cifar = [
2.58e11 * 16.3 * 39,
2.58e11 * 2 * 16.3 * 36,
2.58e11 * 3 * 16.3 * 32,
2.58e11 * 4 * 16.3 * 25,
2.58e11 * 5 * 16.3 * 21,
]
flops_bcnn_acc_cifar = [
5.5e11 * 16.3 * 51,
5.5e11 * 2 * 16.3 * 37,
5.5e11 * 3 * 16.3 * 30,
5.5e11 * 4 * 16.3 * 36,
5.5e11 * 5 * 16.3 * 33,
]
flops_freq_acc_cifar = [
2.58e11 * 16.3 * 7,
2.58e11 * 2 * 16.3 * 4,
2.58e11 * 3 * 16.3 * 4,
2.58e11 * 4 * 16.3 * 3,
2.58e11 * 5 * 16.3 * 3,
]
#=
println("MNIST BCNN 100 FLOPS $(flops_bcnn_100_mnist)")
println("MNIST FREQ 100 FLOPS $(flops_freq_100_mnist)")
println("MNIST BCNN est FLOPS $(flops_bcnn_est_mnist)")
println("MNIST FREQ est FLOPS $(flops_freq_est_mnist)")
println("MNIST BCNN wat FLOPS $(flops_bcnn_wat_mnist)")
println("MNIST FREQ wat FLOPS $(flops_freq_wat_mnist)")
println("MNIST BCNN acc FLOPS $(flops_bcnn_acc_mnist)")
println("MNIST FREQ acc FLOPS $(flops_freq_acc_mnist)")
println("CIFAR BCNN 100 FLOPS $(flops_bcnn_100_cifar)")
println("CIFAR FREQ 100 FLOPS $(flops_freq_100_cifar)")
println("CIFAR BCNN est FLOPS $(flops_bcnn_est_cifar)")
println("CIFAR FREQ est FLOPS $(flops_freq_est_cifar)")
println("CIFAR BCNN wat FLOPS $(flops_bcnn_wat_cifar)")
println("CIFAR FREQ wat FLOPS $(flops_freq_wat_cifar)")
println("CIFAR BCNN acc FLOPS $(flops_bcnn_acc_cifar)")
println("CIFAR FREQ acc FLOPS $(flops_freq_acc_cifar)")
=#
watts_bcnn_100_mnist = [384367.0, 354604.0, 578490.0, 598499.0, 659321.0]
watts_freq_100_mnist = [818766.0, 1.639863e6, 1.826163e6, 1.961074e6, 2.018758e6]
watts_bcnn_est_mnist = [75998.0, 167339.0, 92658.0, 118200.0, 157475.0]
watts_freq_est_mnist = [919386.0, 368970.0, 466917.0, 294907.0, 1.068331e6]
watts_bcnn_wat_mnist = [116516.0, 115347.0, 116225.0, 117989.0, 175601.0]
watts_freq_wat_mnist = [146912.0, 136441.0, 114160.0, 131887.0, 119364.0]
watts_bcnn_acc_mnist = [57569.0, 37169.0, 43488.0, 390792.0, 50428.0]
watts_freq_acc_mnist = [1.332701e6, 498486.0, 532690.0, 1.344363e6, 832005.0]
watts_bcnn_100_cifar = [310789.0, 332406.0, 354631.0, 390792.0, 545941.0]
watts_freq_100_cifar = [1.128901e6, 1.213355e6, 1.29992e6, 1.344363e6, 1.35644e6]
watts_bcnn_est_cifar = [164841.0, 167339.0, 92658.0, 118200.0, 157475.0]
watts_freq_est_cifar = [358373.0, 368970.0, 466917.0, 294907.0, 1.068331e6]
watts_bcnn_wat_cifar = [118621.0, 115347.0, 116225.0, 117989.0, 175601.0]
watts_freq_wat_cifar = [149356.0, 136441.0, 114160.0, 131887.0, 119364.0]
watts_bcnn_acc_cifar = [60415.0, 37169.0, 43488.0, 390792.0, 50428.0]
watts_freq_acc_cifar = [578846.0, 498486.0, 532690.0, 1.344363e6, 832005.0]
#=
println("MNIST BCNN 100 WATTS $(watts_bcnn_100_mnist)")
println("MNIST FREQ 100 WATTS $(watts_freq_100_mnist)")
println("MNIST BCNN est WATTS $(watts_bcnn_est_mnist)")
println("MNIST FREQ est WATTS $(watts_freq_est_mnist)")
println("MNIST BCNN wat WATTS $(watts_bcnn_wat_mnist)")
println("MNIST FREQ wat WATTS $(watts_freq_wat_mnist)")
println("MNIST BCNN acc WATTS $(watts_bcnn_acc_mnist)")
println("MNIST FREQ acc WATTS $(watts_freq_acc_mnist)")
println("CIFAR BCNN 100 WATTS $(watts_bcnn_100_cifar)")
println("CIFAR FREQ 100 WATTS $(watts_freq_100_cifar)")
println("CIFAR BCNN est WATTS $(watts_bcnn_est_cifar)")
println("CIFAR FREQ est WATTS $(watts_freq_est_cifar)")
println("CIFAR BCNN wat WATTS $(watts_bcnn_wat_cifar)")
println("CIFAR FREQ wat WATTS $(watts_freq_wat_cifar)")
println("CIFAR BCNN acc WATTS $(watts_bcnn_acc_cifar)")
println("CIFAR FREQ acc WATTS $(watts_freq_acc_cifar)")
=#
println("MNIST BCNN 100 FLOPS/W $(flops_bcnn_100_mnist ./ watts_bcnn_100_mnist)")
println("MNIST FREQ 100 FLOPS/W $(flops_freq_100_mnist ./ watts_freq_100_mnist)")
println("MNIST BCNN est FLOPS/W $(flops_bcnn_est_mnist ./ watts_bcnn_est_mnist)")
println("MNIST FREQ est FLOPS/W $(flops_freq_est_mnist ./ watts_freq_est_mnist)")
println("MNIST BCNN wat FLOPS/W $(flops_bcnn_wat_mnist ./ watts_bcnn_wat_mnist)")
println("MNIST FREQ wat FLOPS/W $(flops_freq_wat_mnist ./ watts_freq_wat_mnist)")
println("MNIST BCNN acc FLOPS/W $(flops_bcnn_acc_mnist ./ watts_bcnn_acc_mnist)")
println("MNIST FREQ acc FLOPS/W $(flops_freq_acc_mnist ./ watts_freq_acc_mnist)")
println("CIFAR BCNN 100 FLOPS/W $(flops_bcnn_100_cifar ./ watts_bcnn_100_cifar)")
println("CIFAR FREQ 100 FLOPS/W $(flops_freq_100_cifar ./ watts_freq_100_cifar)")
println("CIFAR BCNN est FLOPS/W $(flops_bcnn_est_cifar ./ watts_bcnn_est_cifar)")
println("CIFAR FREQ est FLOPS/W $(flops_freq_est_cifar ./ watts_freq_est_cifar)")
println("CIFAR BCNN wat FLOPS/W $(flops_bcnn_wat_cifar ./ watts_bcnn_wat_cifar)")
println("CIFAR FREQ wat FLOPS/W $(flops_freq_wat_cifar ./ watts_freq_wat_cifar)")
println("CIFAR BCNN acc FLOPS/W $(flops_bcnn_acc_cifar ./ watts_bcnn_acc_cifar)")
println("CIFAR FREQ acc FLOPS/W $(flops_freq_acc_cifar ./ watts_freq_acc_cifar)")
flops_w_bcnn_100_mnist = flops_bcnn_100_mnist ./ watts_bcnn_100_mnist
flops_w_freq_100_mnist = flops_freq_100_mnist ./ watts_freq_100_mnist
flops_w_bcnn_est_mnist = flops_bcnn_est_mnist ./ watts_bcnn_est_mnist
flops_w_freq_est_mnist = flops_freq_est_mnist ./ watts_freq_est_mnist
flops_w_bcnn_wat_mnist = flops_bcnn_wat_mnist ./ watts_bcnn_wat_mnist
flops_w_freq_wat_mnist = flops_freq_wat_mnist ./ watts_freq_wat_mnist
flops_w_bcnn_acc_mnist = flops_bcnn_acc_mnist ./ watts_bcnn_acc_mnist
flops_w_freq_acc_mnist = flops_freq_acc_mnist ./ watts_freq_acc_mnist
flops_w_bcnn_100_cifar = flops_bcnn_100_cifar ./ watts_bcnn_100_cifar
flops_w_freq_100_cifar = flops_freq_100_cifar ./ watts_freq_100_cifar
flops_w_bcnn_est_cifar = flops_bcnn_est_cifar ./ watts_bcnn_est_cifar
flops_w_freq_est_cifar = flops_freq_est_cifar ./ watts_freq_est_cifar
flops_w_bcnn_wat_cifar = flops_bcnn_wat_cifar ./ watts_bcnn_wat_cifar
flops_w_freq_wat_cifar = flops_freq_wat_cifar ./ watts_freq_wat_cifar
flops_w_bcnn_acc_cifar = flops_bcnn_acc_cifar ./ watts_bcnn_acc_cifar
flops_w_freq_acc_cifar = flops_freq_acc_cifar ./ watts_freq_acc_cifar
en_plot = plot(
[
scatter(
x = ["1", "2", "3", "4", "5"],
y = flops_w_freq_100_mnist,
name = "LeNet Exp 1",
marker = attr(symbol = 4, color = "rgb(211,120,000)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = flops_w_freq_est_mnist,
name = "LeNet Exp 2",
marker = attr(symbol = 17, color = "rgb(255,170,017)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = flops_w_freq_wat_mnist,
name = "LeNet Exp 3",
marker = attr(symbol = 2, color = "rgb(255,187,034)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = flops_w_freq_acc_mnist,
name = "LeNet Exp 4",
marker = attr(symbol = 0, color = "rgb(255,204,051)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = flops_w_bcnn_100_mnist,
name = "BCNN Exp 1",
marker = attr(symbol = 4, color = "rgb(055,033,240)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = flops_w_bcnn_est_mnist,
name = "BCNN Exp 2",
marker = attr(symbol = 17, color = "rgb(033,081,240)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = flops_w_bcnn_wat_mnist,
name = "BCNN Exp 3",
marker = attr(symbol = 2, color = "rgb(033,115,240)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = flops_w_bcnn_acc_mnist,
name = "BCNN Exp 4",
marker = attr(symbol = 0, color = "rgb(151,177,255)", line_width = 1.0),
),
],
Layout(
mode = "overlay",
xaxis_tickangle = -45,
yaxis_title_text = "FLOPS/W",
xaxis_title_text = "Size";
xaxis_range = [-1, 5],
xaxis_type = "category",
),
)
savefig(en_plot, "mnist_f_w_exp.png")
en_plot = plot(
[
scatter(
x = ["1", "2", "3", "4", "5"],
y = flops_w_freq_100_cifar,
name = "LeNet Exp 1",
marker = attr(symbol = 4, color = "rgb(211,120,000)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = flops_w_freq_est_cifar,
name = "LeNet Exp 2",
marker = attr(symbol = 17, color = "rgb(255,170,017)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = flops_w_freq_wat_cifar,
name = "LeNet Exp 3",
marker = attr(symbol = 2, color = "rgb(255,187,034)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = flops_w_freq_acc_cifar,
name = "LeNet Exp 4",
marker = attr(symbol = 0, color = "rgb(255,204,051)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = flops_w_bcnn_100_cifar,
name = "BCNN Exp 1",
marker = attr(symbol = 4, color = "rgb(055,033,240)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = flops_w_bcnn_est_cifar,
name = "BCNN Exp 2",
marker = attr(symbol = 17, color = "rgb(033,081,240)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = flops_w_bcnn_wat_cifar,
name = "BCNN Exp 3",
marker = attr(symbol = 2, color = "rgb(033,115,240)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
y = flops_w_bcnn_acc_cifar,
name = "BCNN Exp 4",
marker = attr(symbol = 0, color = "rgb(151,177,255)", line_width = 1.0),
),
],
Layout(
mode = "overlay",
xaxis_tickangle = -45,
yaxis_title_text = "FLOPS/W",
xaxis_title_text = "Size";
xaxis_range = [-1, 5],
xaxis_type = "category",
),
)
savefig(en_plot, "cifar_f_w_exp.png")

390
mean_w_data.jl Executable file
View File

@ -0,0 +1,390 @@
include("aux_func.jl")
using LinearAlgebra
using Statistics
#=
Obtain energy data
=#
#mni_folder_1 = "ini_exp_data/"
mni_folder_100 = "exp_100_epochs/";
cif_folder_100 = "CIFAR_100_epoch/";
mni_folder_acc = "data_bounded/";
cif_folder_acc = "CIFAR_acc_bound/";
mni_folder_wat = "data_budget/";
cif_folder_wat = "CIFAR_energy_bound/";
mni_folder_est = "early_stop_res/";
cif_folder_est = "CIFAR_early_stop/";
bayes_model = "bayes";
freq_model = "freq";
w_type = "watt";
e_type = "exp";
#=
Load GPU data
=#
mni_100_bay_ene = getgpudata(mni_folder_100, bayes_model, w_type, "eo");
mni_100_fre_ene = getgpudata(mni_folder_100, freq_model, w_type, "eo");
cif_100_bay_ene = getgpudata(cif_folder_100, bayes_model, w_type, "eo");
cif_100_fre_ene = getgpudata(cif_folder_100, freq_model, w_type, "eo");
mni_acc_bay_ene = getgpudata(mni_folder_acc, bayes_model, w_type, "eo");
mni_acc_fre_ene = getgpudata(mni_folder_acc, freq_model, w_type, "eo");
cif_acc_bay_ene = getgpudata(cif_folder_acc, bayes_model, w_type, "eo");
cif_acc_fre_ene = getgpudata(cif_folder_acc, freq_model, w_type, "eo");
mni_wat_bay_ene = getgpudata(mni_folder_wat, bayes_model, w_type, "eo");
mni_wat_fre_ene = getgpudata(mni_folder_wat, freq_model, w_type, "eo");
cif_wat_bay_ene = getgpudata(cif_folder_wat, bayes_model, w_type, "eo");
cif_wat_fre_ene = getgpudata(cif_folder_wat, freq_model, w_type, "eo");
mni_est_bay_ene = getgpudata(mni_folder_est, bayes_model, w_type, "eo");
mni_est_fre_ene = getgpudata(mni_folder_est, freq_model, w_type, "eo");
cif_est_bay_ene = getgpudata(cif_folder_est, bayes_model, w_type, "eo");
cif_est_fre_ene = getgpudata(cif_folder_est, freq_model, w_type, "eo");
#=
Define CPU paths
=#
cif = "cifar";
mni = "mnist";
bay = "bayes";
wat = "cpu_watts";
frq = "freq";
ram = "ram_use";
_100 = "100";
acc = "acc";
es = "es";
wbud = "wbud";
#=
Load CPU data
=#
bay_cif_100_ene = readcpudata(cif, bay, wat, _100);
bay_mni_100_ene = readcpudata(mni, bay, wat, _100);
frq_cif_100_ene = readcpudata(cif, frq, wat, _100);
frq_mni_100_ene = readcpudata(mni, frq, wat, _100);
bay_cif_acc_ene = readcpudata(cif, bay, wat, acc);
bay_mni_acc_ene = readcpudata(mni, bay, wat, acc);
frq_cif_acc_ene = readcpudata(cif, frq, wat, acc);
frq_mni_acc_ene = readcpudata(mni, frq, wat, acc);
bay_cif_wbu_ene = readcpudata(cif, bay, wat, wbud);
bay_mni_wbu_ene = readcpudata(mni, bay, wat, wbud);
frq_cif_wbu_ene = readcpudata(cif, frq, wat, wbud);
frq_mni_wbu_ene = readcpudata(mni, frq, wat, wbud);
bay_mni_est_ene = readcpudata(mni, bay, wat, es);
bay_cif_est_ene = readcpudata(cif, bay, wat, es);
frq_cif_est_ene = readcpudata(cif, frq, wat, es);
frq_mni_est_ene = readcpudata(mni, frq, wat, es);
for s = 1:5
bay_cif_acc_ene[s] = round.(getcpuwatt(bay_cif_acc_ene[s]))
bay_cif_est_ene[s] = round.(getcpuwatt(bay_cif_est_ene[s]))
bay_cif_wbu_ene[s] = round.(getcpuwatt(bay_cif_wbu_ene[s]))
bay_cif_100_ene[s] = round.(getcpuwatt(bay_cif_100_ene[s]))
bay_mni_acc_ene[s] = round.(getcpuwatt(bay_mni_acc_ene[s]))
bay_mni_est_ene[s] = round.(getcpuwatt(bay_mni_est_ene[s]))
bay_mni_wbu_ene[s] = round.(getcpuwatt(bay_mni_wbu_ene[s]))
bay_mni_100_ene[s] = round.(getcpuwatt(bay_mni_100_ene[s]))
frq_cif_acc_ene[s] = round.(getcpuwatt(frq_cif_acc_ene[s]))
frq_cif_est_ene[s] = round.(getcpuwatt(frq_cif_est_ene[s]))
frq_cif_wbu_ene[s] = round.(getcpuwatt(frq_cif_wbu_ene[s]))
frq_cif_100_ene[s] = round.(getcpuwatt(frq_cif_100_ene[s]))
frq_mni_acc_ene[s] = round.(getcpuwatt(frq_mni_acc_ene[s]))
frq_mni_est_ene[s] = round.(getcpuwatt(frq_mni_est_ene[s]))
frq_mni_wbu_ene[s] = round.(getcpuwatt(frq_mni_wbu_ene[s]))
frq_mni_100_ene[s] = round.(getcpuwatt(frq_mni_100_ene[s]))
end
for s = 1:5
mni_100_bay_ene[s]["Ene"] = vcat(mni_100_bay_ene[s]["Ene"], bay_mni_100_ene[s])
mni_100_fre_ene[s]["Ene"] = vcat(mni_100_fre_ene[s]["Ene"], frq_mni_100_ene[s])
cif_100_bay_ene[s]["Ene"] = vcat(cif_100_bay_ene[s]["Ene"], bay_cif_100_ene[s])
cif_100_fre_ene[s]["Ene"] = vcat(cif_100_fre_ene[s]["Ene"], frq_cif_100_ene[s])
mni_acc_bay_ene[s]["Ene"] = vcat(mni_acc_bay_ene[s]["Ene"], bay_mni_acc_ene[s])
mni_acc_fre_ene[s]["Ene"] = vcat(mni_acc_fre_ene[s]["Ene"], frq_mni_acc_ene[s])
cif_acc_bay_ene[s]["Ene"] = vcat(cif_acc_bay_ene[s]["Ene"], bay_cif_acc_ene[s])
cif_acc_fre_ene[s]["Ene"] = vcat(cif_acc_fre_ene[s]["Ene"], frq_cif_acc_ene[s])
mni_wat_bay_ene[s]["Ene"] = vcat(mni_wat_bay_ene[s]["Ene"], bay_mni_wbu_ene[s])
mni_wat_fre_ene[s]["Ene"] = vcat(mni_wat_fre_ene[s]["Ene"], frq_mni_wbu_ene[s])
cif_wat_bay_ene[s]["Ene"] = vcat(cif_wat_bay_ene[s]["Ene"], bay_cif_wbu_ene[s])
cif_wat_fre_ene[s]["Ene"] = vcat(cif_wat_fre_ene[s]["Ene"], frq_cif_wbu_ene[s])
mni_est_bay_ene[s]["Ene"] = vcat(mni_est_bay_ene[s]["Ene"], bay_mni_est_ene[s])
mni_est_fre_ene[s]["Ene"] = vcat(mni_est_fre_ene[s]["Ene"], frq_mni_est_ene[s])
cif_est_bay_ene[s]["Ene"] = vcat(cif_est_bay_ene[s]["Ene"], bay_cif_est_ene[s])
cif_est_fre_ene[s]["Ene"] = vcat(cif_est_fre_ene[s]["Ene"], frq_cif_est_ene[s])
end
all_data_ene = Dict(
"mni" => Dict(
1 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
2 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
3 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
4 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
5 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
),
"cif" => Dict(
1 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
2 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
3 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
4 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
5 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
),
)
for s = 1:5
all_data_ene["mni"][s]["100"]["bay"] = mean(mni_100_bay_ene[s]["Ene"])
all_data_ene["mni"][s]["100"]["frq"] = mean(mni_100_fre_ene[s]["Ene"])
all_data_ene["cif"][s]["100"]["bay"] = mean(cif_100_bay_ene[s]["Ene"])
all_data_ene["cif"][s]["100"]["frq"] = mean(cif_100_fre_ene[s]["Ene"])
all_data_ene["mni"][s]["acc"]["bay"] = mean(mni_acc_bay_ene[s]["Ene"])
all_data_ene["mni"][s]["acc"]["frq"] = mean(mni_acc_fre_ene[s]["Ene"])
all_data_ene["cif"][s]["acc"]["bay"] = mean(cif_acc_bay_ene[s]["Ene"])
all_data_ene["cif"][s]["acc"]["frq"] = mean(cif_acc_fre_ene[s]["Ene"])
all_data_ene["mni"][s]["wat"]["bay"] = mean(mni_wat_bay_ene[s]["Ene"])
all_data_ene["mni"][s]["wat"]["frq"] = mean(mni_wat_fre_ene[s]["Ene"])
all_data_ene["cif"][s]["wat"]["bay"] = mean(cif_wat_bay_ene[s]["Ene"])
all_data_ene["cif"][s]["wat"]["frq"] = mean(cif_wat_fre_ene[s]["Ene"])
all_data_ene["mni"][s]["est"]["bay"] = mean(mni_est_bay_ene[s]["Ene"])
all_data_ene["mni"][s]["est"]["frq"] = mean(mni_est_fre_ene[s]["Ene"])
all_data_ene["cif"][s]["est"]["bay"] = mean(cif_est_bay_ene[s]["Ene"])
all_data_ene["cif"][s]["est"]["frq"] = mean(cif_est_fre_ene[s]["Ene"])
end
#=
Load Accuracy data
=#
mni_100_bay_exp = getgpudata(mni_folder_100, bayes_model, e_type, "eo");
mni_100_fre_exp = getgpudata(mni_folder_100, freq_model, e_type, "eo");
cif_100_bay_exp = getgpudata(cif_folder_100, bayes_model, e_type, "eo");
cif_100_fre_exp = getgpudata(cif_folder_100, freq_model, e_type, "eo");
mni_acc_bay_exp = getgpudata(mni_folder_acc, bayes_model, e_type, "eo");
mni_acc_fre_exp = getgpudata(mni_folder_acc, freq_model, e_type, "eo");
cif_acc_bay_exp = getgpudata(cif_folder_acc, bayes_model, e_type, "eo");
cif_acc_fre_exp = getgpudata(cif_folder_acc, freq_model, e_type, "eo");
mni_wat_bay_exp = getgpudata(mni_folder_wat, bayes_model, e_type, "eo");
mni_wat_fre_exp = getgpudata(mni_folder_wat, freq_model, e_type, "eo");
cif_wat_bay_exp = getgpudata(cif_folder_wat, bayes_model, e_type, "eo");
cif_wat_fre_exp = getgpudata(cif_folder_wat, freq_model, e_type, "eo");
mni_est_bay_exp = getgpudata(mni_folder_est, bayes_model, e_type, "eo");
mni_est_fre_exp = getgpudata(mni_folder_est, freq_model, e_type, "eo");
cif_est_bay_exp = getgpudata(cif_folder_est, bayes_model, e_type, "eo");
cif_est_fre_exp = getgpudata(cif_folder_est, freq_model, e_type, "eo");
for i = 1:5
mni_100_bay_exp[i] = expdatatodict(mni_100_bay_exp[i])
mni_100_fre_exp[i] = expdatatodict(mni_100_fre_exp[i])
cif_100_bay_exp[i] = expdatatodict(cif_100_bay_exp[i])
cif_100_fre_exp[i] = expdatatodict(cif_100_fre_exp[i])
mni_acc_bay_exp[i] = expdatatodict(mni_acc_bay_exp[i])
mni_acc_fre_exp[i] = expdatatodict(mni_acc_fre_exp[i])
cif_acc_bay_exp[i] = expdatatodict(cif_acc_bay_exp[i])
cif_acc_fre_exp[i] = expdatatodict(cif_acc_fre_exp[i])
mni_wat_bay_exp[i] = expdatatodict(mni_wat_bay_exp[i])
mni_wat_fre_exp[i] = expdatatodict(mni_wat_fre_exp[i])
cif_wat_bay_exp[i] = expdatatodict(cif_wat_bay_exp[i])
cif_wat_fre_exp[i] = expdatatodict(cif_wat_fre_exp[i])
mni_est_bay_exp[i] = expdatatodict(mni_est_bay_exp[i])
mni_est_fre_exp[i] = expdatatodict(mni_est_fre_exp[i])
cif_est_bay_exp[i] = expdatatodict(cif_est_bay_exp[i])
cif_est_fre_exp[i] = expdatatodict(cif_est_fre_exp[i])
end
all_data_exp = Dict(
"mni" => Dict(
1 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
2 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
3 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
4 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
5 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
),
"cif" => Dict(
1 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
2 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
3 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
4 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
5 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
),
)
for s = 1:5
all_data_exp["mni"][s]["100"]["bay"] =
(mean(mni_100_bay_exp[s]["acc"]) + mean(mni_100_bay_exp[s]["pre"])) / 2
all_data_exp["mni"][s]["100"]["frq"] =
(mean(mni_100_fre_exp[s]["acc"]) + mean(mni_100_fre_exp[s]["pre"])) / 2
all_data_exp["cif"][s]["100"]["bay"] =
(mean(cif_100_bay_exp[s]["acc"]) + mean(cif_100_bay_exp[s]["pre"])) / 2
all_data_ene["cif"][s]["100"]["frq"] =
(mean(cif_100_fre_exp[s]["acc"]) + mean(cif_100_fre_exp[s]["pre"])) / 2
all_data_exp["mni"][s]["acc"]["bay"] =
(mean(mni_acc_bay_exp[s]["acc"]) + mean(mni_acc_bay_exp[s]["pre"])) / 2
all_data_exp["mni"][s]["acc"]["frq"] =
(mean(mni_acc_fre_exp[s]["acc"]) + mean(mni_acc_fre_exp[s]["pre"])) / 2
all_data_exp["cif"][s]["acc"]["bay"] =
(mean(cif_acc_bay_exp[s]["acc"]) + mean(cif_acc_bay_exp[s]["pre"])) / 2
all_data_exp["cif"][s]["acc"]["frq"] =
(mean(cif_acc_fre_exp[s]["acc"]) + mean(cif_acc_fre_exp[s]["pre"])) / 2
all_data_exp["mni"][s]["wat"]["bay"] =
(mean(mni_wat_bay_exp[s]["acc"]) + mean(mni_wat_bay_exp[s]["pre"])) / 2
all_data_exp["mni"][s]["wat"]["frq"] =
(mean(mni_wat_fre_exp[s]["acc"]) + mean(mni_wat_fre_exp[s]["pre"])) / 2
all_data_exp["cif"][s]["wat"]["bay"] =
(mean(cif_wat_bay_exp[s]["acc"]) + mean(cif_wat_bay_exp[s]["pre"])) / 2
all_data_exp["cif"][s]["wat"]["frq"] =
(mean(cif_wat_fre_exp[s]["acc"]) + mean(cif_wat_fre_exp[s]["pre"])) / 2
all_data_exp["mni"][s]["est"]["bay"] =
(mean(mni_est_bay_exp[s]["acc"]) + mean(mni_est_bay_exp[s]["pre"])) / 2
all_data_exp["mni"][s]["est"]["frq"] =
(mean(mni_est_fre_exp[s]["acc"]) + mean(mni_est_fre_exp[s]["pre"])) / 2
all_data_exp["cif"][s]["est"]["bay"] =
(mean(cif_est_bay_exp[s]["acc"]) + mean(cif_est_bay_exp[s]["pre"])) / 2
all_data_exp["cif"][s]["est"]["frq"] =
(mean(cif_est_fre_exp[s]["acc"]) + mean(cif_est_fre_exp[s]["pre"])) / 2
end
#=
# means
=#
(all_data_exp["mni"][5]["100"]["bay"] * 100) / all_data_ene["mni"][5]["100"]["bay"]

1533
memory_compact_processing.jl Executable file
View File

File diff suppressed because it is too large Load Diff

1168
metric_supreme.jl Normal file
View File

File diff suppressed because it is too large Load Diff

285
metric_supreme_graph_50.jl Executable file
View File

@ -0,0 +1,285 @@
include("aux_func.jl")
using PlotlyJS: savefig
using Statistics
using PlotlyJS
using Random
#=
Obtain energy data
=#
#mni_folder_1 = "ini_exp_data/"
mni_folder_100 = "exp_100_epochs/";
cif_folder_100 = "CIFAR_100_epoch/";
mni_folder_acc = "data_bounded/";
cif_folder_acc = "CIFAR_acc_bound/";
mni_folder_wat = "data_budget/";
cif_folder_wat = "CIFAR_energy_bound/";
mni_folder_est = "early_stop_res/";
cif_folder_est = "CIFAR_early_stop/";
bayes_model = "bayes";
freq_model = "freq";
w_type = "watt";
e_type = "exp";
#=
Load GPU data
=#
mni_100_bay_ene = getgpudata(mni_folder_100, bayes_model, w_type, "eo");
mni_100_fre_ene = getgpudata(mni_folder_100, freq_model, w_type, "eo");
cif_100_bay_ene = getgpudata(cif_folder_100, bayes_model, w_type, "eo");
cif_100_fre_ene = getgpudata(cif_folder_100, freq_model, w_type, "eo");
#=
Define CPU paths
=#
cif = "cifar";
mni = "mnist";
bay = "bayes";
wat = "cpu_watts";
frq = "freq";
ram = "ram_use";
_100 = "100";
acc = "acc";
es = "es";
wbud = "wbud";
#=
Load CPU data
=#
bay_cif_100_ene = readcpudata(cif, bay, wat, _100);
bay_mni_100_ene = readcpudata(mni, bay, wat, _100);
frq_cif_100_ene = readcpudata(cif, frq, wat, _100);
frq_mni_100_ene = readcpudata(mni, frq, wat, _100);
for s = 1:5
bay_cif_100_ene[s] = round.(getcpuwatt(bay_cif_100_ene[s]))
bay_mni_100_ene[s] = round.(getcpuwatt(bay_mni_100_ene[s]))
frq_cif_100_ene[s] = round.(getcpuwatt(frq_cif_100_ene[s]))
frq_mni_100_ene[s] = round.(getcpuwatt(frq_mni_100_ene[s]))
end
for s = 1:5
mni_100_bay_ene[s]["Ene"] = vcat(mni_100_bay_ene[s]["Ene"], bay_mni_100_ene[s])
mni_100_fre_ene[s]["Ene"] = vcat(mni_100_fre_ene[s]["Ene"], frq_mni_100_ene[s])
cif_100_bay_ene[s]["Ene"] = vcat(cif_100_bay_ene[s]["Ene"], bay_cif_100_ene[s])
cif_100_fre_ene[s]["Ene"] = vcat(cif_100_fre_ene[s]["Ene"], frq_cif_100_ene[s])
end
ene_data = Dict("mni" => Dict("bay" => Dict(1 => [],2 => [],3 => [],4 => [],5 => []),
"fre" => Dict(1 => [],2 => [],3 => [],4 => [],5 => [])),
"cif" => Dict("bay" => Dict(1 => [],2 => [],3 => [],4 => [],5 => []),
"fre" => Dict(1 => [],2 => [],3 => [],4 => [],5 => [])))
#ene_data = Dict("mni" => Dict("bay" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0),
# "fre" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0)),
# "cif" => Dict("bay" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0),
# "fre" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0)))
for s = 1:5
ene_data["mni"]["bay"][s] = sum.(join_vectors(resize1(makechunks(shuffle(mni_100_bay_ene[s]["Ene"]), 100))[1:50]))
ene_data["mni"]["fre"][s] = sum.(join_vectors(resize1(makechunks(shuffle(mni_100_fre_ene[s]["Ene"]), 100))[1:50]))
ene_data["cif"]["bay"][s] = sum.(join_vectors(resize1(makechunks(shuffle(cif_100_bay_ene[s]["Ene"]), 100))[1:50]))
ene_data["cif"]["fre"][s] = sum.(join_vectors(resize1(makechunks(shuffle(cif_100_fre_ene[s]["Ene"]), 100))[1:50]))
end
#for s = 1:5
# ene_data["mni"]["bay"][s] = sum(mni_100_bay_ene[s]["Ene"]) / 100
# ene_data["mni"]["fre"][s] = sum(mni_100_fre_ene[s]["Ene"]) / 100
# ene_data["cif"]["bay"][s] = sum(cif_100_bay_ene[s]["Ene"]) / 100
# ene_data["cif"]["fre"][s] = sum(cif_100_fre_ene[s]["Ene"]) / 100
#end
#=
Load Accuracy data
=#
mni_100_bay_exp = getgpudata(mni_folder_100, bayes_model, e_type, "eo");
mni_100_fre_exp = getgpudata(mni_folder_100, freq_model, e_type, "eo");
cif_100_bay_exp = getgpudata(cif_folder_100, bayes_model, e_type, "eo");
cif_100_fre_exp = getgpudata(cif_folder_100, freq_model, e_type, "eo");
for i = 1:5
mni_100_bay_exp[i] = expdatatodict(mni_100_bay_exp[i])
mni_100_fre_exp[i] = expdatatodict(mni_100_fre_exp[i])
cif_100_bay_exp[i] = expdatatodict(cif_100_bay_exp[i])
cif_100_fre_exp[i] = expdatatodict(cif_100_fre_exp[i])
end
original_vectors = Dict("LeNet" => Dict("MNIST" => Dict(1 => [],2 => [],3 => [],4 => [],5 => []),
"CIFAR" => Dict(1 => [],2 => [],3 => [],4 => [],5 => [])),
"BCNN" => Dict("MNIST" => Dict(1 => [],2 => [],3 => [],4 => [],5 => []),
"CIFAR" => Dict(1 => [],2 => [],3 => [],4 => [],5 => [])))
for i in 1:5
original_vectors["LeNet"]["MNIST"][i] = (mni_100_fre_exp[i]["acc"][1:50] ./ ene_data["mni"]["fre"][i]) * 100
original_vectors["BCNN"]["MNIST"][i] = (mni_100_bay_exp[i]["acc"][1:50] ./ ene_data["mni"]["bay"][i]) * 100
original_vectors["LeNet"]["CIFAR"][i] = (cif_100_fre_exp[i]["acc"][1:50] ./ ene_data["cif"]["fre"][i]) * 100
original_vectors["BCNN"]["CIFAR"][i] = (cif_100_bay_exp[i]["acc"][1:50] ./ ene_data["cif"]["bay"][i]) * 100
end
en_plot = plot(
[
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors["LeNet"]["MNIST"][1],
name = "LeNet 1",
marker = attr(symbol = 4, color = "rgb(211,120,000)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors["LeNet"]["MNIST"][2],
name = "LeNet 2",
marker = attr(symbol = 17, color = "rgb(255,170,017)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors["LeNet"]["MNIST"][3],
name = "LeNet 3",
marker = attr(symbol = 2, color = "rgb(255,187,034)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors["LeNet"]["MNIST"][4],
name = "LeNet 4",
marker = attr(symbol = 0, color = "rgb(255,204,051)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors["LeNet"]["MNIST"][5],
name = "LeNet 5",
marker = attr(symbol = 4, color = "rgb(255,195,001)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors["BCNN"]["MNIST"][1],
name = "BCNN 1",
marker = attr(symbol = 4, color = "rgb(055,033,240)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors["BCNN"]["MNIST"][2],
name = "BCNN 2",
marker = attr(symbol = 4, color = "rgb(018,094,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors["BCNN"]["MNIST"][3],
name = "BCNN 3",
marker = attr(symbol = 4, color = "rgb(018,126,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors["BCNN"]["MNIST"][4],
name = "BCNN 4",
marker = attr(symbol = 4, color = "rgb(018,150,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors["BCNN"]["MNIST"][5],
name = "BCNN 5",
marker = attr(symbol = 4, color = "rgb(018,167,219)", line_width = 1.0),
),
],
Layout(
mode = "overlay",
xaxis_tickangle = -45,
yaxis_title_text = "Efficency",
xaxis_title_text = "Epoch";
opacity = 0.9,
#xaxis_range = [-1, 5],
#xaxis_type = "category",
),
)
savefig(en_plot, "mnist_efficiency_epoch.png")
en_plot = plot(
[
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors["LeNet"]["CIFAR"][1],
name = "LeNet 1",
marker = attr(symbol = 4, color = "rgb(211,120,000)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors["LeNet"]["CIFAR"][2],
name = "LeNet 2",
marker = attr(symbol = 17, color = "rgb(255,170,017)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors["LeNet"]["CIFAR"][3],
name = "LeNet 3",
marker = attr(symbol = 2, color = "rgb(255,187,034)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors["LeNet"]["CIFAR"][4],
name = "LeNet 4",
marker = attr(symbol = 0, color = "rgb(255,204,051)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors["LeNet"]["CIFAR"][5],
name = "LeNet 5",
marker = attr(symbol = 4, color = "rgb(255,195,001)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors["BCNN"]["CIFAR"][1],
name = "BCNN 1",
marker = attr(symbol = 4, color = "rgb(055,033,240)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors["BCNN"]["CIFAR"][2],
name = "BCNN 2",
marker = attr(symbol = 4, color = "rgb(018,094,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors["BCNN"]["CIFAR"][3],
name = "BCNN 3",
marker = attr(symbol = 4, color = "rgb(018,126,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors["BCNN"]["CIFAR"][4],
name = "BCNN 4",
marker = attr(symbol = 4, color = "rgb(018,150,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors["BCNN"]["CIFAR"][5],
name = "BCNN 5",
marker = attr(symbol = 4, color = "rgb(018,167,219)", line_width = 1.0),
),
],
Layout(
mode = "overlay",
xaxis_tickangle = -45,
yaxis_title_text = "Efficency",
xaxis_title_text = "Epoch";
opacity = 0.9,
#xaxis_range = [-1, 5],
#xaxis_type = "category",
),
)
savefig(en_plot, "cifar_efficiency_epoch.png")

455
metric_supreme_graph_50_acc.jl Executable file
View File

@ -0,0 +1,455 @@
include("aux_func.jl")
using PlotlyJS: savefig
using Statistics
using PlotlyJS
using Random
#=
Obtain energy data
=#
#mni_folder_1 = "ini_exp_data/"
mni_folder_100 = "exp_100_epochs/";
cif_folder_100 = "CIFAR_100_epoch/";
mni_folder_acc = "data_bounded/";
cif_folder_acc = "CIFAR_acc_bound/";
mni_folder_wat = "data_budget/";
cif_folder_wat = "CIFAR_energy_bound/";
mni_folder_est = "early_stop_res/";
cif_folder_est = "CIFAR_early_stop/";
bayes_model = "bayes";
freq_model = "freq";
w_type = "watt";
e_type = "exp";
#=
Load GPU data
=#
mni_100_bay_ene = getgpudata(mni_folder_100, bayes_model, w_type, "eo");
mni_100_fre_ene = getgpudata(mni_folder_100, freq_model, w_type, "eo");
cif_100_bay_ene = getgpudata(cif_folder_100, bayes_model, w_type, "eo");
cif_100_fre_ene = getgpudata(cif_folder_100, freq_model, w_type, "eo");
#=
Define CPU paths
=#
cif = "cifar";
mni = "mnist";
bay = "bayes";
wat = "cpu_watts";
frq = "freq";
ram = "ram_use";
_100 = "100";
acc = "acc";
es = "es";
wbud = "wbud";
#=
Load CPU data
=#
bay_cif_100_ene = readcpudata(cif, bay, wat, _100);
bay_mni_100_ene = readcpudata(mni, bay, wat, _100);
frq_cif_100_ene = readcpudata(cif, frq, wat, _100);
frq_mni_100_ene = readcpudata(mni, frq, wat, _100);
for s = 1:5
bay_cif_100_ene[s] = round.(getcpuwatt(bay_cif_100_ene[s]))
bay_mni_100_ene[s] = round.(getcpuwatt(bay_mni_100_ene[s]))
frq_cif_100_ene[s] = round.(getcpuwatt(frq_cif_100_ene[s]))
frq_mni_100_ene[s] = round.(getcpuwatt(frq_mni_100_ene[s]))
end
for s = 1:5
mni_100_bay_ene[s]["Ene"] = vcat(mni_100_bay_ene[s]["Ene"], bay_mni_100_ene[s])
mni_100_fre_ene[s]["Ene"] = vcat(mni_100_fre_ene[s]["Ene"], frq_mni_100_ene[s])
cif_100_bay_ene[s]["Ene"] = vcat(cif_100_bay_ene[s]["Ene"], bay_cif_100_ene[s])
cif_100_fre_ene[s]["Ene"] = vcat(cif_100_fre_ene[s]["Ene"], frq_cif_100_ene[s])
end
ene_data = Dict("mni" => Dict("bay" => Dict(1 => [],2 => [],3 => [],4 => [],5 => []),
"fre" => Dict(1 => [],2 => [],3 => [],4 => [],5 => [])),
"cif" => Dict("bay" => Dict(1 => [],2 => [],3 => [],4 => [],5 => []),
"fre" => Dict(1 => [],2 => [],3 => [],4 => [],5 => [])))
#ene_data = Dict("mni" => Dict("bay" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0),
# "fre" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0)),
# "cif" => Dict("bay" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0),
# "fre" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0)))
for s = 1:5
ene_data["mni"]["bay"][s] = sum.(join_vectors(resize1(makechunks(shuffle(mni_100_bay_ene[s]["Ene"]), 100))[1:50]))
ene_data["mni"]["fre"][s] = sum.(join_vectors(resize1(makechunks(shuffle(mni_100_fre_ene[s]["Ene"]), 100))[1:50]))
ene_data["cif"]["bay"][s] = sum.(join_vectors(resize1(makechunks(shuffle(cif_100_bay_ene[s]["Ene"]), 100))[1:50]))
ene_data["cif"]["fre"][s] = sum.(join_vectors(resize1(makechunks(shuffle(cif_100_fre_ene[s]["Ene"]), 100))[1:50]))
end
#for s = 1:5
# ene_data["mni"]["bay"][s] = sum(mni_100_bay_ene[s]["Ene"]) / 100
# ene_data["mni"]["fre"][s] = sum(mni_100_fre_ene[s]["Ene"]) / 100
# ene_data["cif"]["bay"][s] = sum(cif_100_bay_ene[s]["Ene"]) / 100
# ene_data["cif"]["fre"][s] = sum(cif_100_fre_ene[s]["Ene"]) / 100
#end
#=
Load Accuracy data
=#
mni_100_bay_exp = getgpudata(mni_folder_100, bayes_model, e_type, "eo");
mni_100_fre_exp = getgpudata(mni_folder_100, freq_model, e_type, "eo");
cif_100_bay_exp = getgpudata(cif_folder_100, bayes_model, e_type, "eo");
cif_100_fre_exp = getgpudata(cif_folder_100, freq_model, e_type, "eo");
for i = 1:5
mni_100_bay_exp[i] = expdatatodict(mni_100_bay_exp[i])
mni_100_fre_exp[i] = expdatatodict(mni_100_fre_exp[i])
cif_100_bay_exp[i] = expdatatodict(cif_100_bay_exp[i])
cif_100_fre_exp[i] = expdatatodict(cif_100_fre_exp[i])
end
original_vectors_acc = Dict("LeNet" => Dict("MNIST" => Dict(1 => [],2 => [],3 => [],4 => [],5 => []),
"CIFAR" => Dict(1 => [],2 => [],3 => [],4 => [],5 => [])),
"BCNN" => Dict("MNIST" => Dict(1 => [],2 => [],3 => [],4 => [],5 => []),
"CIFAR" => Dict(1 => [],2 => [],3 => [],4 => [],5 => [])))
for i in 1:5
original_vectors_acc["LeNet"]["MNIST"][i] = mni_100_fre_exp[i]["acc"][1:50]
original_vectors_acc["BCNN"]["MNIST"][i] = mni_100_bay_exp[i]["acc"][1:50]
original_vectors_acc["LeNet"]["CIFAR"][i] = cif_100_fre_exp[i]["acc"][1:50]
original_vectors_acc["BCNN"]["CIFAR"][i] = cif_100_bay_exp[i]["acc"][1:50]
end
en_plot = plot(
[
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_acc["LeNet"]["MNIST"][1],
name = "LeNet 1",
marker = attr(symbol = 4, color = "rgb(211,120,000)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_acc["LeNet"]["MNIST"][2],
name = "LeNet 2",
marker = attr(symbol = 17, color = "rgb(255,170,017)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_acc["LeNet"]["MNIST"][3],
name = "LeNet 3",
marker = attr(symbol = 2, color = "rgb(255,187,034)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_acc["LeNet"]["MNIST"][4],
name = "LeNet 4",
marker = attr(symbol = 0, color = "rgb(255,204,051)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_acc["LeNet"]["MNIST"][5],
name = "LeNet 5",
marker = attr(symbol = 4, color = "rgb(255,195,001)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_acc["BCNN"]["MNIST"][1],
name = "BCNN 1",
marker = attr(symbol = 4, color = "rgb(055,033,240)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_acc["BCNN"]["MNIST"][2],
name = "BCNN 2",
marker = attr(symbol = 4, color = "rgb(018,094,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_acc["BCNN"]["MNIST"][3],
name = "BCNN 3",
marker = attr(symbol = 4, color = "rgb(018,126,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_acc["BCNN"]["MNIST"][4],
name = "BCNN 4",
marker = attr(symbol = 4, color = "rgb(018,150,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_acc["BCNN"]["MNIST"][5],
name = "BCNN 5",
marker = attr(symbol = 4, color = "rgb(018,167,219)", line_width = 1.0),
),
],
Layout(
mode = "overlay",
xaxis_tickangle = -45,
yaxis_title_text = "Efficency",
xaxis_title_text = "Epoch";
opacity = 0.9,
#xaxis_range = [-1, 5],
yaxis_range = [0, 1],
#xaxis_type = "category",
),
)
savefig(en_plot, "mnist_50_tacc.png")
en_plot = plot(
[
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_acc["LeNet"]["CIFAR"][1],
name = "LeNet 1",
marker = attr(symbol = 4, color = "rgb(211,120,000)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_acc["LeNet"]["CIFAR"][2],
name = "LeNet 2",
marker = attr(symbol = 17, color = "rgb(255,170,017)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_acc["LeNet"]["CIFAR"][3],
name = "LeNet 3",
marker = attr(symbol = 2, color = "rgb(255,187,034)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_acc["LeNet"]["CIFAR"][4],
name = "LeNet 4",
marker = attr(symbol = 0, color = "rgb(255,204,051)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_acc["LeNet"]["CIFAR"][5],
name = "LeNet 5",
marker = attr(symbol = 4, color = "rgb(255,195,001)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_acc["BCNN"]["CIFAR"][1],
name = "BCNN 1",
marker = attr(symbol = 4, color = "rgb(055,033,240)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_acc["BCNN"]["CIFAR"][2],
name = "BCNN 2",
marker = attr(symbol = 4, color = "rgb(018,094,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_acc["BCNN"]["CIFAR"][3],
name = "BCNN 3",
marker = attr(symbol = 4, color = "rgb(018,126,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_acc["BCNN"]["CIFAR"][4],
name = "BCNN 4",
marker = attr(symbol = 4, color = "rgb(018,150,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_acc["BCNN"]["CIFAR"][5],
name = "BCNN 5",
marker = attr(symbol = 4, color = "rgb(018,167,219)", line_width = 1.0),
),
],
Layout(
mode = "overlay",
xaxis_tickangle = -45,
yaxis_title_text = "Efficency",
xaxis_title_text = "Epoch";
opacity = 0.9,
#xaxis_range = [-1, 5],
yaxis_range = [0, 1],
#xaxis_type = "category",
),
)
savefig(en_plot, "cifar_50_tacc.png")
original_vectors_pre = Dict("LeNet" => Dict("MNIST" => Dict(1 => [],2 => [],3 => [],4 => [],5 => []),
"CIFAR" => Dict(1 => [],2 => [],3 => [],4 => [],5 => [])),
"BCNN" => Dict("MNIST" => Dict(1 => [],2 => [],3 => [],4 => [],5 => []),
"CIFAR" => Dict(1 => [],2 => [],3 => [],4 => [],5 => [])))
for i in 1:5
original_vectors_pre["LeNet"]["MNIST"][i] = mni_100_fre_exp[i]["pre"][1:50]
original_vectors_pre["BCNN"]["MNIST"][i] = mni_100_bay_exp[i]["pre"][1:50]
original_vectors_pre["LeNet"]["CIFAR"][i] = cif_100_fre_exp[i]["pre"][1:50]
original_vectors_pre["BCNN"]["CIFAR"][i] = cif_100_bay_exp[i]["pre"][1:50]
end
en_plot = plot(
[
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_pre["LeNet"]["MNIST"][1],
name = "LeNet 1",
marker = attr(symbol = 4, color = "rgb(211,120,000)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_pre["LeNet"]["MNIST"][2],
name = "LeNet 2",
marker = attr(symbol = 17, color = "rgb(255,170,017)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_pre["LeNet"]["MNIST"][3],
name = "LeNet 3",
marker = attr(symbol = 2, color = "rgb(255,187,034)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_pre["LeNet"]["MNIST"][4],
name = "LeNet 4",
marker = attr(symbol = 0, color = "rgb(255,204,051)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_pre["LeNet"]["MNIST"][5],
name = "LeNet 5",
marker = attr(symbol = 4, color = "rgb(255,195,001)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_pre["BCNN"]["MNIST"][1],
name = "BCNN 1",
marker = attr(symbol = 4, color = "rgb(055,033,240)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_pre["BCNN"]["MNIST"][2],
name = "BCNN 2",
marker = attr(symbol = 4, color = "rgb(018,094,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_pre["BCNN"]["MNIST"][3],
name = "BCNN 3",
marker = attr(symbol = 4, color = "rgb(018,126,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_pre["BCNN"]["MNIST"][4],
name = "BCNN 4",
marker = attr(symbol = 4, color = "rgb(018,150,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_pre["BCNN"]["MNIST"][5],
name = "BCNN 5",
marker = attr(symbol = 4, color = "rgb(018,167,219)", line_width = 1.0),
),
],
Layout(
mode = "overlay",
xaxis_tickangle = -45,
yaxis_title_text = "Efficency",
xaxis_title_text = "Epoch";
opacity = 0.9,
#xaxis_range = [-1, 5],
yaxis_range = [0, 1],
#xaxis_type = "category",
),
)
savefig(en_plot, "mnist_50_tpre.png")
en_plot = plot(
[
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_pre["LeNet"]["CIFAR"][1],
name = "LeNet 1",
marker = attr(symbol = 4, color = "rgb(211,120,000)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_pre["LeNet"]["CIFAR"][2],
name = "LeNet 2",
marker = attr(symbol = 17, color = "rgb(255,170,017)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_pre["LeNet"]["CIFAR"][3],
name = "LeNet 3",
marker = attr(symbol = 2, color = "rgb(255,187,034)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_pre["LeNet"]["CIFAR"][4],
name = "LeNet 4",
marker = attr(symbol = 0, color = "rgb(255,204,051)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_pre["LeNet"]["CIFAR"][5],
name = "LeNet 5",
marker = attr(symbol = 4, color = "rgb(255,195,001)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_pre["BCNN"]["CIFAR"][1],
name = "BCNN 1",
marker = attr(symbol = 4, color = "rgb(055,033,240)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_pre["BCNN"]["CIFAR"][2],
name = "BCNN 2",
marker = attr(symbol = 4, color = "rgb(018,094,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_pre["BCNN"]["CIFAR"][3],
name = "BCNN 3",
marker = attr(symbol = 4, color = "rgb(018,126,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_pre["BCNN"]["CIFAR"][4],
name = "BCNN 4",
marker = attr(symbol = 4, color = "rgb(018,150,219)", line_width = 1.0),
),
scatter(
#x = ["1", "2", "3", "4", "5"],
y = original_vectors_pre["BCNN"]["CIFAR"][5],
name = "BCNN 5",
marker = attr(symbol = 4, color = "rgb(018,167,219)", line_width = 1.0),
),
],
Layout(
mode = "overlay",
xaxis_tickangle = -45,
yaxis_title_text = "Efficency",
xaxis_title_text = "Epoch";
opacity = 0.9,
#xaxis_range = [-1, 5],
yaxis_range = [0, 1],
#xaxis_type = "category",
),
)
savefig(en_plot, "cifar_50_tpre.png")

157
metric_supreme_graph_acc.jl Executable file
View File

@ -0,0 +1,157 @@
include("aux_func.jl")
using PlotlyJS: savefig
using Statistics
using PlotlyJS
using Random
using Printf
#=
Obtain energy data
=#
#mni_folder_1 = "ini_exp_data/"
mni_folder_100 = "exp_100_epochs/";
cif_folder_100 = "CIFAR_100_epoch/";
mni_folder_acc = "data_bounded/";
cif_folder_acc = "CIFAR_acc_bound/";
mni_folder_wat = "data_budget/";
cif_folder_wat = "CIFAR_energy_bound/";
mni_folder_est = "early_stop_res/";
cif_folder_est = "CIFAR_early_stop/";
bayes_model = "bayes";
freq_model = "freq";
w_type = "watt";
e_type = "exp";
#=
Load GPU data
=#
mni_100_bay_ene = getgpudata(mni_folder_100, bayes_model, w_type, "eo");
mni_100_fre_ene = getgpudata(mni_folder_100, freq_model, w_type, "eo");
cif_100_bay_ene = getgpudata(cif_folder_100, bayes_model, w_type, "eo");
cif_100_fre_ene = getgpudata(cif_folder_100, freq_model, w_type, "eo");
#=
Define CPU paths
=#
cif = "cifar";
mni = "mnist";
bay = "bayes";
wat = "cpu_watts";
frq = "freq";
ram = "ram_use";
_100 = "100";
acc = "acc";
es = "es";
wbud = "wbud";
#=
Load CPU data
=#
bay_cif_100_ene = readcpudata(cif, bay, wat, _100);
bay_mni_100_ene = readcpudata(mni, bay, wat, _100);
frq_cif_100_ene = readcpudata(cif, frq, wat, _100);
frq_mni_100_ene = readcpudata(mni, frq, wat, _100);
for s = 1:5
bay_cif_100_ene[s] = round.(getcpuwatt(bay_cif_100_ene[s]))
bay_mni_100_ene[s] = round.(getcpuwatt(bay_mni_100_ene[s]))
frq_cif_100_ene[s] = round.(getcpuwatt(frq_cif_100_ene[s]))
frq_mni_100_ene[s] = round.(getcpuwatt(frq_mni_100_ene[s]))
end
for s = 1:5
mni_100_bay_ene[s]["Ene"] = vcat(mni_100_bay_ene[s]["Ene"], bay_mni_100_ene[s])
mni_100_fre_ene[s]["Ene"] = vcat(mni_100_fre_ene[s]["Ene"], frq_mni_100_ene[s])
cif_100_bay_ene[s]["Ene"] = vcat(cif_100_bay_ene[s]["Ene"], bay_cif_100_ene[s])
cif_100_fre_ene[s]["Ene"] = vcat(cif_100_fre_ene[s]["Ene"], frq_cif_100_ene[s])
end
ene_data = Dict("mni" => Dict("bay" => Dict(1 => [],2 => [],3 => [],4 => [],5 => []),
"fre" => Dict(1 => [],2 => [],3 => [],4 => [],5 => [])),
"cif" => Dict("bay" => Dict(1 => [],2 => [],3 => [],4 => [],5 => []),
"fre" => Dict(1 => [],2 => [],3 => [],4 => [],5 => [])))
#ene_data = Dict("mni" => Dict("bay" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0),
# "fre" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0)),
# "cif" => Dict("bay" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0),
# "fre" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0)))
for s = 1:5
ene_data["mni"]["bay"][s] = sum.(join_vectors(resize1(makechunks(shuffle(mni_100_bay_ene[s]["Ene"]), 100))[1:100]))
ene_data["mni"]["fre"][s] = sum.(join_vectors(resize1(makechunks(shuffle(mni_100_fre_ene[s]["Ene"]), 100))[1:100]))
ene_data["cif"]["bay"][s] = sum.(join_vectors(resize1(makechunks(shuffle(cif_100_bay_ene[s]["Ene"]), 100))[1:100]))
ene_data["cif"]["fre"][s] = sum.(join_vectors(resize1(makechunks(shuffle(cif_100_fre_ene[s]["Ene"]), 100))[1:100]))
end
#for s = 1:5
# ene_data["mni"]["bay"][s] = sum(mni_100_bay_ene[s]["Ene"]) / 100
# ene_data["mni"]["fre"][s] = sum(mni_100_fre_ene[s]["Ene"]) / 100
# ene_data["cif"]["bay"][s] = sum(cif_100_bay_ene[s]["Ene"]) / 100
# ene_data["cif"]["fre"][s] = sum(cif_100_fre_ene[s]["Ene"]) / 100
#end
#=
Load Accuracy data
=#
mni_100_bay_exp = getgpudata(mni_folder_100, bayes_model, e_type, "eo");
mni_100_fre_exp = getgpudata(mni_folder_100, freq_model, e_type, "eo");
cif_100_bay_exp = getgpudata(cif_folder_100, bayes_model, e_type, "eo");
cif_100_fre_exp = getgpudata(cif_folder_100, freq_model, e_type, "eo");
for i = 1:5
mni_100_bay_exp[i] = expdatatodict(mni_100_bay_exp[i])
mni_100_fre_exp[i] = expdatatodict(mni_100_fre_exp[i])
cif_100_bay_exp[i] = expdatatodict(cif_100_bay_exp[i])
cif_100_fre_exp[i] = expdatatodict(cif_100_fre_exp[i])
end
original_vectors_acc = Dict("LeNet" => Dict("MNIST" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0),
"CIFAR" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0)),
"BCNN" => Dict("MNIST" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0),
"CIFAR" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0)))
for i in 1:5
original_vectors_acc["LeNet"]["MNIST"][i] = mean(mni_100_fre_exp[i]["acc"][1:100])
@printf "LeNet-5 on MNIST acc size %i Metric: %-.8f\n" i original_vectors_acc["LeNet"]["MNIST"][i]
original_vectors_acc["BCNN"]["MNIST"][i] = mean(mni_100_bay_exp[i]["acc"][1:100])
@printf "BCNN-5 on MNIST acc size %i Metric: %-.8f\n" i original_vectors_acc["BCNN"]["MNIST"][i]
original_vectors_acc["LeNet"]["CIFAR"][i] = mean(cif_100_fre_exp[i]["acc"][1:100])
@printf "LeNet-5 on CIFAR acc size %i Metric: %-.8f\n" i original_vectors_acc["LeNet"]["CIFAR"][i]
original_vectors_acc["BCNN"]["CIFAR"][i] = mean(cif_100_bay_exp[i]["acc"][1:100])
@printf "BCNN-5 on CIFAR acc size %i Metric: %-.8f\n" i original_vectors_acc["BCNN"]["CIFAR"][i]
end
println("\n\n\n")
original_vectors_pre = Dict("LeNet" => Dict("MNIST" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0),
"CIFAR" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0)),
"BCNN" => Dict("MNIST" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0),
"CIFAR" => Dict(1 => 0.0,2 => 0.0,3 => 0.0,4 => 0.0,5 => 0.0)))
for i in 1:5
original_vectors_pre["LeNet"]["MNIST"][i] = mean(mni_100_fre_exp[i]["pre"][1:100])
@printf "LeNet-5 on MNIST pre size %i Metric: %-.8f\n" i original_vectors_pre["LeNet"]["MNIST"][i]
original_vectors_pre["BCNN"]["MNIST"][i] = mean(mni_100_bay_exp[i]["pre"][1:100])
@printf "BCNN-5 on MNIST pre size %i Metric: %-.8f\n" i original_vectors_pre["BCNN"]["MNIST"][i]
original_vectors_pre["LeNet"]["CIFAR"][i] = mean(cif_100_fre_exp[i]["pre"][1:100])
@printf "LeNet-5 on CIFAR pre size %i Metric: %-.8f\n" i original_vectors_pre["LeNet"]["CIFAR"][i]
original_vectors_pre["BCNN"]["CIFAR"][i] = mean(cif_100_bay_exp[i]["pre"][1:100])
@printf "BCNN-5 on CIFAR pre size %i Metric: %-.8f\n" i original_vectors_pre["BCNN"]["CIFAR"][i]
end

386
metric_supreme_graphs.jl Normal file
View File

@ -0,0 +1,386 @@
using PlotlyJS
using PlotlyJS: savefig
using Statistics: mean, std
using DataFrames
include("aux_func.jl")
data = load_pickle("efficiency_per_size_sum.pkl");
#all_data_ene["mni"][s]["50"]["bay"]
data_type = ["mni", "cif"]
model_type = ["bay", "frq"]
experiment_type = ["100", "est", "acc", "wat"]
experiment_100=["50","50","50","50","50"]
experiment_est=["est","est","est","est","est"]
experiment_wat=["wat","wat","wat","wat","wat"]
experiment_acc=["acc","acc","acc","acc","acc"]
model_bayes = ["BCNN","BCNN","BCNN","BCNN","BCNN"]
model_lenet = ["LeNet","LeNet","LeNet","LeNet","LeNet"]
data_mnist = ["MNIST","MNIST","MNIST","MNIST","MNIST"]
data_cifar = ["CIFAR","CIFAR","CIFAR","CIFAR","CIFAR"]
sizes = [1,2,3,4,5]
# MNIST 100 efficency
bayes_100_mnist_eff = []
for s = 1:5
push!(
bayes_100_mnist_eff,
data[model_type[1]][data_type[1]][experiment_type[1]][string(s)],
)
end
bayes_100_mnist_eff = DataFrame(Efficiency=bayes_100_mnist_eff,Model=model_bayes,Experiment=experiment_100,Dataset=data_mnist,Size=sizes)
freqs_100_mnist_eff = []
for s = 1:5
push!(
freqs_100_mnist_eff,
data[model_type[2]][data_type[1]][experiment_type[1]][string(s)],
)
end
freqs_100_mnist_eff = DataFrame(Efficiency=freqs_100_mnist_eff,Model=model_lenet,Experiment=experiment_100,Dataset=data_mnist,Size=sizes)
# MNIST est efficency
bayes_est_mnist_eff = []
for s = 1:5
push!(
bayes_est_mnist_eff,
data[model_type[1]][data_type[1]][experiment_type[2]][string(s)],
)
end
bayes_est_mnist_eff = DataFrame(Efficiency=bayes_est_mnist_eff,Model=model_bayes,Experiment=experiment_est,Dataset=data_mnist,Size=sizes)
freqs_est_mnist_eff = []
for s = 1:5
push!(
freqs_est_mnist_eff,
data[model_type[2]][data_type[1]][experiment_type[2]][string(s)],
)
end
freqs_est_mnist_eff = DataFrame(Efficiency=freqs_est_mnist_eff,Model=model_lenet,Experiment=experiment_est,Dataset=data_mnist,Size=sizes)
# MNIST wat efficency
bayes_wat_mnist_eff = []
for s = 1:5
push!(
bayes_wat_mnist_eff,
data[model_type[1]][data_type[1]][experiment_type[4]][string(s)],
)
end
bayes_wat_mnist_eff = DataFrame(Efficiency=bayes_wat_mnist_eff,Model=model_bayes,Experiment=experiment_wat,Dataset=data_mnist,Size=sizes)
freqs_wat_mnist_eff = []
for s = 1:5
push!(
freqs_wat_mnist_eff,
data[model_type[2]][data_type[1]][experiment_type[4]][string(s)],
)
end
freqs_wat_mnist_eff = DataFrame(Efficiency=freqs_wat_mnist_eff,Model=model_lenet,Experiment=experiment_wat,Dataset=data_mnist,Size=sizes)
# MNIST acc efficency
bayes_acc_mnist_eff = []
for s = 1:5
push!(
bayes_acc_mnist_eff,
data[model_type[1]][data_type[1]][experiment_type[3]][string(s)],
)
end
bayes_acc_mnist_eff = DataFrame(Efficiency=bayes_acc_mnist_eff,Model=model_bayes,Experiment=experiment_acc,Dataset=data_mnist,Size=sizes)
freqs_acc_mnist_eff = []
for s = 1:5
push!(
freqs_acc_mnist_eff,
data[model_type[2]][data_type[1]][experiment_type[3]][string(s)],
)
end
freqs_acc_mnist_eff = DataFrame(Efficiency=freqs_acc_mnist_eff,Model=model_lenet,Experiment=experiment_acc,Dataset=data_mnist,Size=sizes)
# CIFAR 100 efficency
bayes_100_cifar_eff = []
for s = 1:5
push!(
bayes_100_cifar_eff,
data[model_type[1]][data_type[2]][experiment_type[1]][string(s)],
)
end
bayes_100_cifar_eff = DataFrame(Efficiency=bayes_100_cifar_eff,Model=model_bayes,Experiment=experiment_100,Dataset=data_cifar,Size=sizes)
#for i = 1:5
# t_std = std(bayes_100_cifar_eff)
# if (bayes_100_cifar_eff[i] > 2 * t_std) || (bayes_100_cifar_eff[i] < 2 * t_std)
# bayes_100_cifar_eff[i] = mean(bayes_100_cifar_eff)
# end
#end
freqs_100_cifar_eff = []
for s = 1:5
push!(
freqs_100_cifar_eff,
data[model_type[2]][data_type[2]][experiment_type[1]][string(s)],
)
end
freqs_100_cifar_eff = DataFrame(Efficiency=freqs_100_cifar_eff,Model=model_lenet,Experiment=experiment_100,Dataset=data_cifar,Size=sizes)
#for i = 1:5
# t_std = std(freqs_100_cifar_eff)
# if (freqs_100_cifar_eff[i] > 2 * t_std) || (freqs_100_cifar_eff[i] < 2 * t_std)
# freqs_100_cifar_eff[i] = mean(freqs_100_cifar_eff)
# end
#end
# CIFAR est efficency
bayes_est_cifar_eff = []
for s = 1:5
push!(
bayes_est_cifar_eff,
data[model_type[1]][data_type[2]][experiment_type[2]][string(s)],
)
end
bayes_est_cifar_eff = DataFrame(Efficiency=bayes_est_cifar_eff,Model=model_bayes,Experiment=experiment_est,Dataset=data_cifar,Size=sizes)
freqs_est_cifar_eff = []
for s = 1:5
push!(
freqs_est_cifar_eff,
data[model_type[2]][data_type[2]][experiment_type[2]][string(s)],
)
end
freqs_est_cifar_eff = DataFrame(Efficiency=freqs_est_cifar_eff,Model=model_lenet,Experiment=experiment_est,Dataset=data_cifar,Size=sizes)
# CIFAR wat efficency
bayes_wat_cifar_eff = []
for s = 1:5
push!(
bayes_wat_cifar_eff,
data[model_type[1]][data_type[2]][experiment_type[4]][string(s)],
)
end
bayes_wat_cifar_eff = DataFrame(Efficiency=bayes_wat_cifar_eff,Model=model_bayes,Experiment=experiment_wat,Dataset=data_cifar,Size=sizes)
freqs_wat_cifar_eff = []
for s = 1:5
push!(
freqs_wat_cifar_eff,
data[model_type[2]][data_type[2]][experiment_type[4]][string(s)],
)
end
freqs_wat_cifar_eff = DataFrame(Efficiency=freqs_wat_cifar_eff,Model=model_lenet,Experiment=experiment_wat,Dataset=data_cifar,Size=sizes)
# CIFAR acc efficency
bayes_acc_cifar_eff = []
for s = 1:5
push!(
bayes_acc_cifar_eff,
data[model_type[1]][data_type[2]][experiment_type[3]][string(s)],
)
end
bayes_acc_cifar_eff = DataFrame(Efficiency=bayes_acc_cifar_eff,Model=model_bayes,Experiment=experiment_acc,Dataset=data_cifar,Size=sizes)
freqs_acc_cifar_eff = []
for s = 1:5
push!(
freqs_acc_cifar_eff,
data[model_type[2]][data_type[2]][experiment_type[3]][string(s)],
)
end
freqs_acc_cifar_eff = DataFrame(Efficiency=freqs_acc_cifar_eff,Model=model_lenet,Experiment=experiment_acc,Dataset=data_cifar,Size=sizes)
mnist_dataframe = vcat(bayes_100_mnist_eff,freqs_100_mnist_eff,bayes_est_mnist_eff,freqs_est_mnist_eff,bayes_acc_mnist_eff,freqs_acc_mnist_eff,bayes_wat_mnist_eff,freqs_wat_mnist_eff)
cifar_dataframe = vcat(bayes_100_cifar_eff,freqs_100_cifar_eff,bayes_est_cifar_eff,freqs_est_cifar_eff,bayes_acc_cifar_eff,freqs_acc_cifar_eff,bayes_wat_cifar_eff,freqs_wat_cifar_eff)
#avg_rate_bcnn = (-8.266684252643054e-5 * 1000)
#avg_rate_fcnn = (0.00022035677966088333 * 1000)
#=
en_plot = plot(
[
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_fcnn .* freqs_100_mnist_eff),
y = freqs_100_mnist_eff.Efficiency,
name = "LeNet 100",
marker = attr(symbol = 4, color = "rgb(211,120,000)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_fcnn .* freqs_est_mnist_eff),
y = freqs_est_mnist_eff.Efficiency,
name = "LeNet est",
marker = attr(symbol = 17, color = "rgb(255,170,017)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_fcnn .* freqs_wat_mnist_eff),
y = freqs_wat_mnist_eff.Efficiency,
name = "LeNet wat",
marker = attr(symbol = 2, color = "rgb(255,187,034)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_fcnn .* freqs_acc_mnist_eff),
y = freqs_acc_mnist_eff.Efficiency,
name = "LeNet acc",
marker = attr(symbol = 0, color = "rgb(255,204,051)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_bcnn .* bayes_100_mnist_eff),
y = bayes_100_mnist_eff.Efficiency,
name = "BCNN 100",
marker = attr(symbol = 4, color = "rgb(055,033,240)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_bcnn .* bayes_est_mnist_eff),
y = bayes_est_mnist_eff.Efficiency,
name = "BCNN est",
marker = attr(symbol = 17, color = "rgb(033,081,240)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_bcnn .* bayes_wat_mnist_eff),
y = bayes_wat_mnist_eff.Efficiency,
name = "BCNN wat",
marker = attr(symbol = 2, color = "rgb(033,115,240)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_bcnn .* bayes_acc_mnist_eff),
y = bayes_acc_mnist_eff.Efficiency,
name = "BCNN acc",
marker = attr(symbol = 0, color = "rgb(151,177,255)", line_width = 1.0),
),
],
Layout(
mode = "overlay",
xaxis_tickangle = -45,
yaxis_title_text = "Efficiency",
xaxis_title_text = "Size";
xaxis_range = [-1, 5],
xaxis_type = "category",
),
)
savefig(en_plot, "mnist_eff_exp_sum.png")
en_plot = plot(
[
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_fcnn .* freqs_100_cifar_eff),
y = freqs_100_cifar_eff.Efficiency,
name = "LeNet 100",
marker = attr(symbol = 4, color = "rgb(211,120,000)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_fcnn .* freqs_est_cifar_eff),
y = freqs_est_cifar_eff.Efficiency,
name = "LeNet est",
marker = attr(symbol = 17, color = "rgb(255,170,017)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_fcnn .* freqs_wat_cifar_eff),
y = freqs_wat_cifar_eff.Efficiency,
name = "LeNet wat",
marker = attr(symbol = 2, color = "rgb(255,187,034)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_fcnn .* freqs_acc_cifar_eff),
y = freqs_acc_cifar_eff.Efficiency,
name = "LeNet acc",
marker = attr(symbol = 0, color = "rgb(255,204,051)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_bcnn .* bayes_100_cifar_eff),
y = bayes_100_cifar_eff.Efficiency,
name = "BCNN 100",
marker = attr(symbol = 4, color = "rgb(055,033,240)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_bcnn .* bayes_est_cifar_eff),
y = bayes_est_cifar_eff.Efficiency,
name = "BCNN est",
marker = attr(symbol = 17, color = "rgb(033,081,240)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_bcnn .* bayes_wat_cifar_eff),
y = bayes_wat_cifar_eff.Efficiency,
name = "BCNN wat",
marker = attr(symbol = 2, color = "rgb(033,115,240)", line_width = 1.0),
),
scatter(
x = ["1", "2", "3", "4", "5"],
#y = (-avg_rate_bcnn .* bayes_acc_cifar_eff),
y = bayes_acc_cifar_eff.Efficiency,
name = "BCNN acc",
marker = attr(symbol = 0, color = "rgb(151,177,255)", line_width = 1.0),
),
],
Layout(
mode = "overlay",
xaxis_tickangle = -45,
yaxis_title_text = "Efficiency",
xaxis_title_text = "Size";
xaxis_range = [-1, 5],
xaxis_type = "category",
),
)
savefig(en_plot, "cifar_eff_exp_sum.png")
=#
#en_plot = plot(mnist_dataframe, x=:Experiment, y=:Efficiency, boxpoints="all", kind="box")
en_plot = plot(
mnist_dataframe, x=:Experiment, y=:Efficiency,kind="scatter",mode="markers",color=:Model,
quartilemethod="exclusive",
marker=attr(size=:Size, sizeref=maximum(mnist_dataframe.Size) / (10^2), sizemode="area")
#marker=attr(size=:Size, sizeref=0.1, sizemode="area")
)
savefig(en_plot, "mnist_scatter_size.png")
en_plot = plot(
mnist_dataframe, x=:Experiment, y=:Efficiency,kind="box", boxpoints="all",color=:Model,
quartilemethod="exclusive",
marker=attr(size=:Size, sizeref=maximum(mnist_dataframe.Size) / (10^2), sizemode="area")
#marker=attr(size=:Size, sizeref=0.1, sizemode="area")
)
savefig(en_plot, "mnist_box_size.png")
en_plot = plot(
cifar_dataframe, x=:Experiment, y=:Efficiency,kind="scatter",mode="markers",color=:Model,
quartilemethod="exclusive",
marker=attr(size=:Size, sizeref=maximum(mnist_dataframe.Size) / (10^2), sizemode="area")
#marker=attr(size=:Size, sizeref=0.1, sizemode="area")
)
savefig(en_plot, "cifar_scatter_size.png")
en_plot = plot(
cifar_dataframe, x=:Experiment, y=:Efficiency,kind="box", boxpoints="all",color=:Model,
quartilemethod="exclusive",
marker=attr(size=:Size, sizeref=maximum(mnist_dataframe.Size) / (10^2), sizemode="area")
#marker=attr(size=:Size, sizeref=0.1, sizemode="area")
)
savefig(en_plot, "cifar_box_size.png")

663
one_epoch_flop_proc.jl Executable file
View File

@ -0,0 +1,663 @@
include("aux_func.jl")
using LinearAlgebra
using Statistics
using PlotlyJS
#using HDF5
folder = "ini_exp_data/"
# Getting the BCNN double precision data
dp_bcnn = read("$(folder)dp_bcnn.txt", String);
dp_bcnn = split(dp_bcnn, "\n")
bool_all = occursin.(r"smsp", dp_bcnn)
dp_bcnn_all = dp_bcnn[bool_all]
dp_bcnn_all = strip.(dp_bcnn_all)
bool_add = occursin.(r"dadd", dp_bcnn_all)
dp_bcnn_add = dp_bcnn_all[bool_add]
dp_bcnn_add = split.(dp_bcnn_add, " ")
dp_bcnn_add = filter.(x -> x != "", dp_bcnn_add)
temp = []
for arr in dp_bcnn_add
append!(temp, parse(Int64, replace(arr[3], r"," => "")))
end
dp_bcnn_add = temp[:]
bool_fma = occursin.(r"dfma", dp_bcnn_all)
dp_bcnn_fma = dp_bcnn_all[bool_fma]
dp_bcnn_fma = split.(dp_bcnn_fma, " ")
dp_bcnn_fma = filter.(x -> x != "", dp_bcnn_fma)
temp = []
for arr in dp_bcnn_fma
append!(temp, parse(Int64, replace(arr[3], r"," => "")))
end
dp_bcnn_fma = temp[:]
bool_mul = occursin.(r"dmul", dp_bcnn_all)
dp_bcnn_mul = dp_bcnn_all[bool_mul]
dp_bcnn_mul = split.(dp_bcnn_mul, " ")
dp_bcnn_mul = filter.(x -> x != "", dp_bcnn_mul)
temp = []
for arr in dp_bcnn_mul
append!(temp, parse(Int64, replace(arr[3], r"," => "")))
end
dp_bcnn_mul = temp[:]
#println("Total bcnn double precision FLOPS: $(sum(dp_bcnn_fma)*2 + sum(dp_bcnn_add) + sum(dp_bcnn_mul))")
# Getting the BCNN half precision data
hp_bcnn = read("$(folder)hp_bcnn.txt", String);
hp_bcnn = split(hp_bcnn, "\n")
bool_all = occursin.(r"smsp", hp_bcnn)
hp_bcnn_all = hp_bcnn[bool_all]
hp_bcnn_all = strip.(hp_bcnn_all)
bool_add = occursin.(r"hadd", hp_bcnn_all)
hp_bcnn_add = hp_bcnn_all[bool_add]
hp_bcnn_add = split.(hp_bcnn_add, " ")
hp_bcnn_add = filter.(x -> x != "", hp_bcnn_add)
temp = []
for arr in hp_bcnn_add
append!(temp, parse(Int64, replace(arr[3], r"," => "")))
end
hp_bcnn_add = temp[:]
bool_fma = occursin.(r"hfma", hp_bcnn_all)
hp_bcnn_fma = hp_bcnn_all[bool_fma]
hp_bcnn_fma = split.(hp_bcnn_fma, " ")
hp_bcnn_fma = filter.(x -> x != "", hp_bcnn_fma)
temp = []
for arr in hp_bcnn_fma
append!(temp, parse(Int64, replace(arr[3], r"," => "")))
end
hp_bcnn_fma = temp[:]
bool_mul = occursin.(r"hmul", hp_bcnn_all)
hp_bcnn_mul = hp_bcnn_all[bool_mul]
hp_bcnn_mul = split.(hp_bcnn_mul, " ")
hp_bcnn_mul = filter.(x -> x != "", hp_bcnn_mul)
temp = []
for arr in hp_bcnn_mul
append!(temp, parse(Int64, replace(arr[3], r"," => "")))
end
hp_bcnn_mul = temp[:]
#println("Total bcnn half precision FLOPS: $(sum(hp_bcnn_fma)*2 + sum(hp_bcnn_add) + sum(hp_bcnn_mul))")
# Getting the BCNN single precision data
sp_bcnn = read("$(folder)sp_bcnn.txt", String);
sp_bcnn = split(sp_bcnn, "\n")
bool_all = occursin.(r"smsp", sp_bcnn)
sp_bcnn_all = sp_bcnn[bool_all]
sp_bcnn_all = strip.(sp_bcnn_all)
bool_add = occursin.(r"fadd", sp_bcnn_all)
sp_bcnn_add = sp_bcnn_all[bool_add]
sp_bcnn_add = split.(sp_bcnn_add, " ")
sp_bcnn_add = filter.(x -> x != "", sp_bcnn_add)
temp = []
for arr in sp_bcnn_add
append!(temp, parse(Int64, replace(arr[3], r"," => "")))
end
sp_bcnn_add = temp[:]
bool_fma = occursin.(r"ffma", sp_bcnn_all)
sp_bcnn_fma = sp_bcnn_all[bool_fma]
sp_bcnn_fma = split.(sp_bcnn_fma, " ")
sp_bcnn_fma = filter.(x -> x != "", sp_bcnn_fma)
temp = []
for arr in sp_bcnn_fma
append!(temp, parse(Int64, replace(arr[3], r"," => "")))
end
sp_bcnn_fma = temp[:]
bool_mul = occursin.(r"fmul", sp_bcnn_all)
sp_bcnn_mul = sp_bcnn_all[bool_mul]
sp_bcnn_mul = split.(sp_bcnn_mul, " ")
sp_bcnn_mul = filter.(x -> x != "", sp_bcnn_mul)
temp = []
for arr in sp_bcnn_mul
append!(temp, parse(Int64, replace(arr[3], r"," => "")))
end
sp_bcnn_mul = temp[:]
#println("Total bcnn single precision FLOPS: $(sum(sp_bcnn_fma)*2 + sum(sp_bcnn_add) + sum(sp_bcnn_mul))")
# Getting the BCNN memory data
mem_bcnn = read("$(folder)memrw_bcnn.txt", String);
mem_bcnn = split(mem_bcnn, "\n")
bool_all = occursin.(r"dram", mem_bcnn)
mem_bcnn_all = mem_bcnn[bool_all]
mem_bcnn_all = strip.(mem_bcnn_all)
bool_read = occursin.(r"read", mem_bcnn_all)
mem_bcnn_read = mem_bcnn_all[bool_read]
mem_bcnn_read = split.(mem_bcnn_read, " ")
mem_bcnn_read = filter.(x -> x != "", mem_bcnn_read)
bool_write = occursin.(r"write", mem_bcnn_all)
mem_bcnn_write = mem_bcnn_all[bool_write]
mem_bcnn_write = split.(mem_bcnn_write, " ")
mem_bcnn_write = filter.(x -> x != "", mem_bcnn_write)
mem_bcnn_read = converttobyte(mem_bcnn_read)
mem_bcnn_write = converttobyte(mem_bcnn_write)
#println("Total BCNN memory writes $(sum(mem_bcnn_write)/1048576.0) Mb")
#println("Total BCNN memory reads $(sum(mem_bcnn_read)/1048576.0) Mb")
# Getting the CIFAR BCNN memory data
cif_mem_bcnn = read("$(folder)cif_memrw_bcnn.txt", String);
cif_mem_bcnn = split(cif_mem_bcnn, "\n")
cif_bool_all = occursin.(r"dram", cif_mem_bcnn)
cif_mem_bcnn_all = cif_mem_bcnn[cif_bool_all]
cif_mem_bcnn_all = strip.(cif_mem_bcnn_all)
cif_bool_read = occursin.(r"read", cif_mem_bcnn_all)
cif_mem_bcnn_read = cif_mem_bcnn_all[cif_bool_read]
cif_mem_bcnn_read = split.(cif_mem_bcnn_read, " ")
cif_mem_bcnn_read = filter.(x -> x != "", cif_mem_bcnn_read)
cif_bool_write = occursin.(r"write", cif_mem_bcnn_all)
cif_mem_bcnn_write = cif_mem_bcnn_all[cif_bool_write]
cif_mem_bcnn_write = split.(cif_mem_bcnn_write, " ")
cif_mem_bcnn_write = filter.(x -> x != "", cif_mem_bcnn_write)
cif_mem_bcnn_read = converttobyte(cif_mem_bcnn_read)
cif_mem_bcnn_write = converttobyte(cif_mem_bcnn_write)
#println("Total CIFAR BCNN memory writes $(sum(cif_mem_bcnn_write)/1048576.0) Mb")
#println("Total CIFAR BCNN memory reads $(sum(cif_mem_bcnn_read)/1048576.0) Mb")
# Getting the CNN double precision data
dp_fcnn = read("$(folder)dp_fcnn.txt", String);
dp_fcnn = split(dp_fcnn, "\n")
bool_all = occursin.(r"smsp", dp_fcnn)
dp_fcnn_all = dp_fcnn[bool_all]
dp_fcnn_all = strip.(dp_fcnn_all)
bool_add = occursin.(r"dadd", dp_fcnn_all)
dp_fcnn_add = dp_fcnn_all[bool_add]
dp_fcnn_add = split.(dp_fcnn_add, " ")
dp_fcnn_add = filter.(x -> x != "", dp_fcnn_add)
temp = []
for arr in dp_fcnn_add
append!(temp, parse(Int64, replace(arr[3], r"," => "")))
end
dp_fcnn_add = temp[:]
bool_fma = occursin.(r"dfma", dp_fcnn_all)
dp_fcnn_fma = dp_fcnn_all[bool_fma]
dp_fcnn_fma = split.(dp_fcnn_fma, " ")
dp_fcnn_fma = filter.(x -> x != "", dp_fcnn_fma)
temp = []
for arr in dp_fcnn_fma
append!(temp, parse(Int64, replace(arr[3], r"," => "")))
end
dp_fcnn_fma = temp[:]
bool_mul = occursin.(r"dmul", dp_fcnn_all)
dp_fcnn_mul = dp_fcnn_all[bool_mul]
dp_fcnn_mul = split.(dp_fcnn_mul, " ")
dp_fcnn_mul = filter.(x -> x != "", dp_fcnn_mul)
temp = []
for arr in dp_fcnn_mul
append!(temp, parse(Int64, replace(arr[3], r"," => "")))
end
dp_fcnn_mul = temp[:]
#println("Total cnn double precision FLOPS: $(sum(dp_fcnn_fma)*2 + sum(dp_fcnn_add) + sum(dp_fcnn_mul))")
# Getting the CNN half precision data
hp_fcnn = read("$(folder)hp_fcnn.txt", String);
hp_fcnn = split(hp_fcnn, "\n")
bool_all = occursin.(r"smsp", hp_fcnn)
hp_fcnn_all = hp_fcnn[bool_all]
hp_fcnn_all = strip.(hp_fcnn_all)
bool_add = occursin.(r"hadd", hp_fcnn_all)
hp_fcnn_add = hp_fcnn_all[bool_add]
hp_fcnn_add = split.(hp_fcnn_add, " ")
hp_fcnn_add = filter.(x -> x != "", hp_fcnn_add)
temp = []
for arr in hp_fcnn_add
append!(temp, parse(Int64, replace(arr[3], r"," => "")))
end
hp_fcnn_add = temp[:]
bool_fma = occursin.(r"hfma", hp_fcnn_all)
hp_fcnn_fma = hp_fcnn_all[bool_fma]
hp_fcnn_fma = split.(hp_fcnn_fma, " ")
hp_fcnn_fma = filter.(x -> x != "", hp_fcnn_fma)
temp = []
for arr in hp_fcnn_fma
append!(temp, parse(Int64, replace(arr[3], r"," => "")))
end
hp_fcnn_fma = temp[:]
bool_mul = occursin.(r"hmul", hp_fcnn_all)
hp_fcnn_mul = hp_fcnn_all[bool_mul]
hp_fcnn_mul = split.(hp_fcnn_mul, " ")
hp_fcnn_mul = filter.(x -> x != "", hp_fcnn_mul)
temp = []
for arr in hp_fcnn_mul
append!(temp, parse(Int64, replace(arr[3], r"," => "")))
end
hp_fcnn_mul = temp[:]
#println("Total cnn half precision FLOPS: $(sum(hp_fcnn_fma)*2 + sum(hp_fcnn_add) + sum(hp_fcnn_mul))")
# Getting the CNN single precision data
sp_fcnn = read("$(folder)sp_fcnn.txt", String);
sp_fcnn = split(sp_fcnn, "\n")
bool_all = occursin.(r"smsp", sp_fcnn)
sp_fcnn_all = sp_fcnn[bool_all]
sp_fcnn_all = strip.(sp_fcnn_all)
bool_add = occursin.(r"fadd", sp_fcnn_all)
sp_fcnn_add = sp_fcnn_all[bool_add]
sp_fcnn_add = split.(sp_fcnn_add, " ")
sp_fcnn_add = filter.(x -> x != "", sp_fcnn_add)
temp = []
for arr in sp_fcnn_add
append!(temp, parse(Int64, replace(arr[3], r"," => "")))
end
sp_fcnn_add = temp[:]
bool_fma = occursin.(r"ffma", sp_fcnn_all)
sp_fcnn_fma = sp_fcnn_all[bool_fma]
sp_fcnn_fma = split.(sp_fcnn_fma, " ")
sp_fcnn_fma = filter.(x -> x != "", sp_fcnn_fma)
temp = []
for arr in sp_fcnn_fma
append!(temp, parse(Int64, replace(arr[3], r"," => "")))
end
sp_fcnn_fma = temp[:]
bool_mul = occursin.(r"fmul", sp_fcnn_all)
sp_fcnn_mul = sp_fcnn_all[bool_mul]
sp_fcnn_mul = split.(sp_fcnn_mul, " ")
sp_fcnn_mul = filter.(x -> x != "", sp_fcnn_mul)
temp = []
for arr in sp_fcnn_mul
append!(temp, parse(Int64, replace(arr[3], r"," => "")))
end
sp_fcnn_mul = temp[:]
#println("Total cnn single precision FLOPS: $(sum(sp_fcnn_fma)*2 + sum(sp_fcnn_add) + sum(sp_fcnn_mul))")
# Getting the CNN memory data
mem_fcnn = read("$(folder)memrw_fcnn.txt", String);
mem_fcnn = split(mem_fcnn, "\n")
bool_all = occursin.(r"dram", mem_fcnn)
mem_fcnn_all = mem_fcnn[bool_all]
mem_fcnn_all = strip.(mem_fcnn_all)
bool_read = occursin.(r"read", mem_fcnn_all)
mem_fcnn_read = mem_fcnn_all[bool_read]
mem_fcnn_read = split.(mem_fcnn_read, " ")
mem_fcnn_read = filter.(x -> x != "", mem_fcnn_read)
bool_write = occursin.(r"write", mem_fcnn_all)
mem_fcnn_write = mem_fcnn_all[bool_write]
mem_fcnn_write = split.(mem_fcnn_write, " ")
mem_fcnn_write = filter.(x -> x != "", mem_fcnn_write)
mem_fcnn_read = converttobyte(mem_fcnn_read)
mem_fcnn_write = converttobyte(mem_fcnn_write)
#println("Total CNN memory writes $(sum(mem_fcnn_write)/1048576.0) Mb")
#println("Total CNN memory reads $(sum(mem_fcnn_read)/1048576.0) Mb")
# Getting the CIFAR CNN memory data
cif_mem_fcnn = read("$(folder)cif_memrw_fcnn.txt", String);
cif_mem_fcnn = split(cif_mem_fcnn, "\n")
cif_bool_all = occursin.(r"dram", cif_mem_fcnn)
cif_mem_fcnn_all = cif_mem_fcnn[cif_bool_all]
cif_mem_fcnn_all = strip.(cif_mem_fcnn_all)
cif_bool_read = occursin.(r"read", cif_mem_fcnn_all)
cif_mem_fcnn_read = cif_mem_fcnn_all[cif_bool_read]
cif_mem_fcnn_read = split.(cif_mem_fcnn_read, " ")
cif_mem_fcnn_read = filter.(x -> x != "", cif_mem_fcnn_read)
cif_bool_write = occursin.(r"write", cif_mem_fcnn_all)
cif_mem_fcnn_write = cif_mem_fcnn_all[cif_bool_write]
cif_mem_fcnn_write = split.(cif_mem_fcnn_write, " ")
cif_mem_fcnn_write = filter.(x -> x != "", cif_mem_fcnn_write)
cif_mem_fcnn_read = converttobyte(cif_mem_fcnn_read)
cif_mem_fcnn_write = converttobyte(cif_mem_fcnn_write)
#println("Total CIFAR CNN memory writes $(sum(cif_mem_fcnn_write)/1048576.0) Mb")
#println("Total CIFAR CNN memory reads $(sum(cif_mem_fcnn_read)/1048576.0) Mb")
# Getting the CIFAR BCNN single precision data
cif_sp_bcnn = read("$(folder)cif_sp_bcnn.txt", String);
cif_sp_bcnn = split(cif_sp_bcnn, "\n")
bool_all = occursin.(r"smsp", cif_sp_bcnn)
cif_sp_bcnn_all = cif_sp_bcnn[bool_all]
cif_sp_bcnn_all = strip.(cif_sp_bcnn_all)
bool_add = occursin.(r"fadd", cif_sp_bcnn_all)
cif_sp_bcnn_add = cif_sp_bcnn_all[bool_add]
cif_sp_bcnn_add = split.(cif_sp_bcnn_add, " ")
cif_sp_bcnn_add = filter.(x -> x != "", cif_sp_bcnn_add)
temp = []
for arr in cif_sp_bcnn_add
append!(temp, parse(Int64, replace(arr[3], r"," => "")))
end
cif_sp_bcnn_add = temp[:]
bool_fma = occursin.(r"ffma", cif_sp_bcnn_all)
cif_sp_bcnn_fma = cif_sp_bcnn_all[bool_fma]
cif_sp_bcnn_fma = split.(cif_sp_bcnn_fma, " ")
cif_sp_bcnn_fma = filter.(x -> x != "", cif_sp_bcnn_fma)
temp = []
for arr in cif_sp_bcnn_fma
append!(temp, parse(Int64, replace(arr[3], r"," => "")))
end
cif_sp_bcnn_fma = temp[:]
bool_mul = occursin.(r"fmul", cif_sp_bcnn_all)
cif_sp_bcnn_mul = cif_sp_bcnn_all[bool_mul]
cif_sp_bcnn_mul = split.(cif_sp_bcnn_mul, " ")
cif_sp_bcnn_mul = filter.(x -> x != "", cif_sp_bcnn_mul)
temp = []
for arr in cif_sp_bcnn_mul
append!(temp, parse(Int64, replace(arr[3], r"," => "")))
end
cif_sp_bcnn_mul = temp[:]
#println("Total bcnn CIFAR single precision FLOPS: $(sum(cif_sp_bcnn_fma)*2 + sum(cif_sp_bcnn_add) + sum(cif_sp_bcnn_mul))")
# Getting the CIFAR CNN single precision data
cif_sp_fcnn = read("$(folder)cif_sp_fcnn.txt", String);
cif_sp_fcnn = split(cif_sp_fcnn, "\n")
bool_all = occursin.(r"smsp", cif_sp_fcnn)
cif_sp_fcnn_all = cif_sp_fcnn[bool_all]
cif_sp_fcnn_all = strip.(cif_sp_fcnn_all)
bool_add = occursin.(r"fadd", cif_sp_fcnn_all)
cif_sp_fcnn_add = cif_sp_fcnn_all[bool_add]
cif_sp_fcnn_add = split.(cif_sp_fcnn_add, " ")
cif_sp_fcnn_add = filter.(x -> x != "", cif_sp_fcnn_add)
temp = []
for arr in cif_sp_fcnn_add
append!(temp, parse(Int64, replace(arr[3], r"," => "")))
end
cif_sp_fcnn_add = temp[:]
bool_fma = occursin.(r"ffma", cif_sp_fcnn_all)
cif_sp_fcnn_fma = cif_sp_fcnn_all[bool_fma]
cif_sp_fcnn_fma = split.(cif_sp_fcnn_fma, " ")
cif_sp_fcnn_fma = filter.(x -> x != "", cif_sp_fcnn_fma)
temp = []
for arr in cif_sp_fcnn_fma
append!(temp, parse(Int64, replace(arr[3], r"," => "")))
end
cif_sp_fcnn_fma = temp[:]
bool_mul = occursin.(r"fmul", cif_sp_fcnn_all)
cif_sp_fcnn_mul = cif_sp_fcnn_all[bool_mul]
cif_sp_fcnn_mul = split.(cif_sp_fcnn_mul, " ")
cif_sp_fcnn_mul = filter.(x -> x != "", cif_sp_fcnn_mul)
temp = []
for arr in cif_sp_fcnn_mul
append!(temp, parse(Int64, replace(arr[3], r"," => "")))
end
cif_sp_fcnn_mul = temp[:]
#println("Total cnn CIFAR single precision FLOPS: $(sum(sp_fcnn_fma)*2 + sum(sp_fcnn_add) + sum(sp_fcnn_mul))")
#println("\n\n\n\n\n\n")
# Data moves
mov_pj = 0.00125
# Turing arch values
add_μj = 0.0293
mul_μj = 0.2933
fma_μj = 0.8340
# Intel arch values
add_pj = 0.000111
mul_pj = 0.000164
fma_pj = 0.000464
tot_mem_bcnn_w_mb = sum(mem_bcnn_write) #sum((1/norm(mem_bcnn_write)).*mem_bcnn_write) /100
tot_mem_bcnn_r_mb = sum(mem_bcnn_read) #sum((1/norm(mem_bcnn_read)).*mem_bcnn_read) /100
tot_mem_fcnn_w_mb = sum(mem_fcnn_write) #sum((1/norm(mem_fcnn_write)).*mem_fcnn_write) /100
tot_mem_fcnn_r_mb = sum(mem_fcnn_read) #sum((1/norm(mem_fcnn_read)).*mem_fcnn_read) /100
rate_mem_bcnn = (tot_mem_bcnn_w_mb / tot_mem_bcnn_r_mb) * mov_pj
rate_mem_fcnn = (tot_mem_fcnn_w_mb / tot_mem_fcnn_r_mb) * mov_pj
cif_tot_mem_bcnn_w_mb = sum(cif_mem_bcnn_write) #sum((1/norm(cif_mem_bcnn_write)).*cif_mem_bcnn_write) /100
cif_tot_mem_bcnn_r_mb = sum(cif_mem_bcnn_read) #sum((1/norm(cif_mem_bcnn_read)).*cif_mem_bcnn_read) /100
cif_tot_mem_fcnn_w_mb = sum(cif_mem_fcnn_write) #sum((1/norm(cif_mem_fcnn_write)).*cif_mem_fcnn_write) /100
cif_tot_mem_fcnn_r_mb = sum(cif_mem_fcnn_read) #sum((1/norm(cif_mem_fcnn_read)).*cif_mem_fcnn_read) /100
cif_rate_mem_bcnn = (cif_tot_mem_bcnn_w_mb / cif_tot_mem_bcnn_r_mb) * mov_pj
cif_rate_mem_fcnn = (cif_tot_mem_fcnn_w_mb / cif_tot_mem_fcnn_r_mb) * mov_pj
tot_bcnn_add = sum(sp_bcnn_add) #sum((1/norm(sp_bcnn_add)).*sp_bcnn_add) /100
tot_bcnn_mul = sum(sp_bcnn_mul) #sum((1/norm(sp_bcnn_mul)).*sp_bcnn_mul) /100
tot_bcnn_fma = sum(sp_bcnn_fma) #sum((1/norm(sp_bcnn_fma)).*sp_bcnn_fma) /100
tot_fcnn_add = sum(sp_fcnn_add) #sum((1/norm(sp_fcnn_add)).*sp_fcnn_add) /100
tot_fcnn_mul = sum(sp_fcnn_mul) #sum((1/norm(sp_fcnn_mul)).*sp_fcnn_mul) /100
tot_fcnn_fma = sum(sp_fcnn_fma) #sum((1/norm(sp_fcnn_fma)).*sp_fcnn_fma) /100
rat_bcnn_add_fma = sum(sp_bcnn_add) / sum(sp_bcnn_fma)
rat_bcnn_mul_fma = sum(sp_bcnn_mul) / sum(sp_bcnn_fma)
rat_fcnn_add_fma = sum(sp_fcnn_add) / sum(sp_fcnn_fma)
rat_fcnn_mul_fma = sum(sp_fcnn_mul) / sum(sp_fcnn_fma)
#=
println("Ratio BCNN ADD/FMA: $(rat_bcnn_add_fma)")
println("Ratio BCNN MUL/FMA: $(rat_bcnn_mul_fma)")
println("Ratio FCNN ADD/FMA: $(rat_fcnn_add_fma)")
println("Ratio FCNN MUL/FMA: $(rat_fcnn_mul_fma)")
=#
cif_tot_bcnn_add = sum(cif_sp_bcnn_add) #sum((1/norm(cif_sp_bcnn_add)).*cif_sp_bcnn_add) /100
cif_tot_bcnn_mul = sum(cif_sp_bcnn_mul) #sum((1/norm(cif_sp_bcnn_mul)).*cif_sp_bcnn_mul) /100
cif_tot_bcnn_fma = sum(cif_sp_bcnn_fma) #sum((1/norm(cif_sp_bcnn_fma)).*cif_sp_bcnn_fma) /100
cif_tot_fcnn_add = sum(cif_sp_fcnn_add) #sum((1/norm(cif_sp_fcnn_add)).*cif_sp_fcnn_add) /100
cif_tot_fcnn_mul = sum(cif_sp_fcnn_mul) #sum((1/norm(cif_sp_fcnn_mul)).*cif_sp_fcnn_mul) /100
cif_tot_fcnn_fma = sum(cif_sp_fcnn_fma) #sum((1/norm(cif_sp_fcnn_fma)).*cif_sp_fcnn_fma) /100
cif_rat_bcnn_add_fma = sum(cif_sp_bcnn_add) / sum(cif_sp_bcnn_fma)
cif_rat_bcnn_mul_fma = sum(cif_sp_bcnn_fma) / sum(cif_sp_bcnn_fma)
cif_rat_fcnn_add_fma = sum(cif_sp_fcnn_add) / sum(cif_sp_fcnn_fma)
cif_rat_fcnn_mul_fma = sum(cif_sp_fcnn_mul) / sum(cif_sp_fcnn_fma)
#=
println("Ratio CIFAR BCNN ADD/FMA: $(cif_rat_bcnn_add_fma)")
println("Ratio CIFAR BCNN MUL/FMA: $(cif_rat_bcnn_mul_fma)")
println("Ratio CIFAR FCNN ADD/FMA: $(cif_rat_fcnn_add_fma)")
println("Ratio CIFAR FCNN MUL/FMA: $(cif_rat_fcnn_mul_fma)")
=#
#add_ratio_bccn = rat_bcnn_add_fma - cif_rat_bcnn_add_fma #* (1 - (add_μj + add_pj))
#mul_ratio_bccn = rat_bcnn_mul_fma - cif_rat_bcnn_mul_fma #* (1 - (mul_μj + mul_pj))
#add_ratio_fccn = rat_fcnn_add_fma - cif_rat_fcnn_add_fma #* (1 - (add_μj + add_pj))
#mul_ratio_fccn = rat_fcnn_mul_fma - cif_rat_fcnn_mul_fma #* (1 - (mul_μj + mul_pj))
#=
println("Memory rate of change BCNN: $(sum(mem_bcnn_write)/sum(mem_bcnn_read))")
println("Memory rate of change FCNN: $(sum(mem_fcnn_write)/sum(mem_fcnn_read))")
println("Memory rate of change CIFAR BCNN: $(sum(cif_mem_bcnn_write)/sum(cif_mem_bcnn_read))")
println("Memory rate of change CIFAR FCNN: $(sum(cif_mem_fcnn_write)/sum(cif_mem_fcnn_read))")
=#
#println("Rate of change ADD BCNN: $(add_ratio_bccn)")
#println("Rate of change MUL BCNN: $(mul_ratio_bccn)")
#println("Rate of change ADD FCNN: $(add_ratio_fccn)")
#println("Rate of change MUL FCNN: $(mul_ratio_fccn)")
#avg_rate_bcnn =
# ((add_ratio_bccn + mul_ratio_bccn) / 2) * (((add_μj + add_pj) + (mul_μj + mul_pj)) / 2)
#avg_rate_fcnn =
# ((add_ratio_fccn + mul_ratio_fccn) / 2) * (((add_μj + add_pj) + (mul_μj + mul_pj)) / 2)
#println("Avg rate BCNN: $(avg_rate_bcnn)")
#println("Avg rate FCNN: $(avg_rate_fcnn)")
# get total scaled
bcnn_mnist_vector = [tot_bcnn_add,tot_bcnn_mul,tot_bcnn_fma,tot_mem_bcnn_r_mb,tot_mem_bcnn_w_mb]
bcnn_cifar_vector = [cif_tot_bcnn_add,cif_tot_bcnn_mul,cif_tot_bcnn_fma,cif_tot_mem_bcnn_r_mb,cif_tot_mem_bcnn_w_mb]
fcnn_mnist_vector = [tot_fcnn_add,tot_fcnn_mul,tot_fcnn_fma,tot_mem_bcnn_r_mb,tot_mem_bcnn_w_mb]
fcnn_cifar_vector = [cif_tot_fcnn_add,cif_tot_fcnn_mul,cif_tot_fcnn_fma,cif_tot_mem_fcnn_r_mb,cif_tot_mem_fcnn_w_mb]
#bcnn_mnist_vector_max = max(bcnn_mnist_vector...)
#bcnn_cifar_vector_max = max(bcnn_cifar_vector...)
#
#fcnn_mnist_vector_max = max(fcnn_mnist_vector...)
#fcnn_cifar_vector_max = max(fcnn_cifar_vector...)
#
#bcnn_mnist_vector = bcnn_mnist_vector ./ (bcnn_mnist_vector_max * 1)
#bcnn_cifar_vector = bcnn_cifar_vector ./ (bcnn_cifar_vector_max * 1)
#
#fcnn_mnist_vector = fcnn_mnist_vector ./ (fcnn_mnist_vector_max * 1)
#fcnn_cifar_vector = fcnn_cifar_vector ./ (fcnn_cifar_vector_max * 1)
en_plot = plot(
[
bar(
y=bcnn_mnist_vector,
x=["ADD","MUL","FMA","MRE","MWR"],
#cumulative_enabled=false,
name="MNIST",
#marker_color="#eb98b5",
marker_color = "rgb(055,033,240)",
opacity=0.75
),
bar(
y=bcnn_cifar_vector,
#y=fcnn_mnist_vector,
x=["ADD", "MUL","FMA","MRE","MWR"],
#cumulative_enabled=false,
name="CIFAR",
#marker_color="#330C73",
marker_color = "rgb(211,120,000)",
opacity=0.75
)
],
#Layout(title="BCNN", xaxis_title="Operations", yaxis_title="Total")#, yaxis_range=[0,1])
Layout(xaxis_title="Operations", yaxis_title="Total")#, yaxis_range=[0,1])
#Layout(xaxis_title="Operations", yaxis_title="Total", yaxis_range=[0,1])
)
savefig(en_plot, "bcnn_oper.png")
#savefig(en_plot, "mnist_oper.png")
en_plot = plot(
[
bar(
y=fcnn_mnist_vector,
#y=bcnn_cifar_vector,
x=["ADD","MUL","FMA","MRE","MWR"],
#cumulative_enabled=false,
name="MNIST",
#marker_color="#eb98b5",
marker_color = "rgb(055,033,240)",
opacity=0.75
),
bar(
y=fcnn_cifar_vector,
x=["ADD", "MUL","FMA","MRE","MWR"],
#cumulative_enabled=false,
name="CIFAR",
#marker_color="#330C73",
marker_color = "rgb(211,120,000)",
opacity=0.75
)
],
#Layout(title="FCNN", xaxis_title="Operations", yaxis_title="Total")#, yaxis_range=[0,1])
Layout(xaxis_title="Operations", yaxis_title="Total")#, yaxis_range=[0,1])
#Layout(xaxis_title="Operations", yaxis_title="Total", yaxis_range=[0,1])
)
savefig(en_plot, "fcnn_oper.png")
#savefig(en_plot, "cifar_oper.png")
#=
amount = 500
# plots
#mem_read_plot = plot(mem_bcnn_read[1:amount],title="First $(amount) memory reads",label="BNN reads");plot!(mem_fcnn_read[1:amount],label="CNN reads")
#savefig(mem_read_plot,"mem_read.png")
mem_read_plot = plot(mem_bcnn_read[1:amount],label="BNN reads");plot!(mem_fcnn_read[1:amount],label="CNN reads")
savefig(mem_read_plot,"mem_read.png")
#mem_write_plot = plot(mem_bcnn_write[1:amount],title="First $(amount) memory writes",label="BNN writes");plot!(mem_fcnn_write[1:amount],label="CNN writes")
#savefig(mem_write_plot,"mem_write.png")
mem_write_plot = plot(mem_bcnn_write[1:amount],label="BNN writes");plot!(mem_fcnn_write[1:amount],label="CNN writes")
savefig(mem_write_plot,"mem_write.png")
#add_plot = plot(sp_bcnn_add[1:amount],title="First $(amount) single precision ADD operations",label="BNN ADD");plot!(sp_fcnn_add[1:amount],label="CNN ADD")
#savefig(add_plot,"add_plot.png")
add_plot = plot(sp_bcnn_add[1:amount],label="BNN ADD");plot!(sp_fcnn_add[1:amount],label="CNN ADD")
savefig(add_plot,"add_plot.png")
#mul_plot = plot(sp_bcnn_mul[1:amount],title="First $(amount) single precision MUL operations",label="BNN MUL");plot!(sp_fcnn_mul[1:amount],label="CNN MUL")
#savefig(mul_plot,"mul_plot.png")
mul_plot = plot(sp_bcnn_mul[1:amount],label="BNN MUL");plot!(sp_fcnn_mul[1:amount],label="CNN MUL")
savefig(mul_plot,"mul_plot.png")
#fma_plot = plot(sp_bcnn_fma[1:amount],title="First $(amount) single precision FMA operations",label="BNN FMA");plot!(sp_fcnn_fma[1:amount],label="CNN FMA")
#savefig(fma_plot,"fma_plot.png")
fma_plot = plot(sp_bcnn_fma[1:amount],label="BNN FMA");plot!(sp_fcnn_fma[1:amount],label="CNN FMA")
savefig(fma_plot,"fma_plot.png")
#mem_bcnn_read=convert.(Float64,mem_bcnn_read)
#mem_bcnn_write=convert.(Float64,mem_bcnn_write)
#mem_fcnn_read=convert.(Float64,mem_fcnn_read)
#mem_fcnn_write=convert.(Float64,mem_fcnn_write)
#sp_bcnn_add=convert.(Float64,sp_bcnn_add)
#sp_bcnn_fma=convert.(Float64,sp_bcnn_fma)
#sp_bcnn_mul=convert.(Float64,sp_bcnn_mul)
#sp_fcnn_add=convert.(Float64,sp_fcnn_add)
#sp_fcnn_fma=convert.(Float64,sp_fcnn_fma)
#sp_fcnn_mul=convert.(Float64,sp_fcnn_mul)
#h5write("mem_bcnn_read.h5", "mem_bcnn_read", mem_bcnn_read)
#h5write("mem_bcnn_write.h5", "mem_bcnn_write", mem_bcnn_write)
#h5write("mem_fcnn_read.h5", "mem_fcnn_read", mem_fcnn_read)
#h5write("mem_fcnn_write.h5", "mem_fcnn_write", mem_fcnn_write)
#h5write("sp_bcnn_add.h5", "sp_bcnn_add", sp_bcnn_add)
#h5write("sp_bcnn_fma.h5", "sp_bcnn_fma", sp_bcnn_fma)
#h5write("sp_bcnn_mul.h5", "sp_bcnn_mul", sp_bcnn_mul)
#h5write("sp_fcnn_add.h5", "sp_fcnn_add", sp_fcnn_add)
#h5write("sp_fcnn_fma.h5", "sp_fcnn_fma", sp_fcnn_fma)
#h5write("sp_fcnn_mul.h5", "sp_fcnn_mul", sp_fcnn_mul)
=#

151
one_epoch_watt_proc.jl Executable file
View File

@ -0,0 +1,151 @@
#using HDF5
include("aux_func.jl")
#using Plots
#using StatsPlots
using Statistics
using LinearAlgebra
using PlotlyJS, Random, StatsBase
using PlotlyJS: savefig
folder = "ini_exp_data/"
#fcnn = h5open("fcnn_watt_data.h5", "r") do file
# read(file,"fcnn")
#end
#bcnn = h5open("bcnn_watt_data.h5", "r") do file
# read(file,"bcnn")
#end
fcnn = load_pickle("$(folder)fcnn_wattdata.pkl")
bcnn = load_pickle("$(folder)bcnn_wattdata.pkl")
bcnn_w = removewatt(bcnn[:, 1])
mem_bcnn = removewatt(bcnn[:, 3])
#bcnn_ch = bcnn_w .> 28
#bcnn_start = findall(x -> x==1, bcnn_ch)[1]
#bcnn_end = findall(x -> x==1, bcnn_ch)[end]
#bcnn_w_tot = bcnn_w[bcnn_start:bcnn_end]
#bcnn_tot = sum(bcnn_w_tot)
#mean_bcnn_w = mean(bcnn_w_tot)
#bcnn_s = size(bcnn_w_tot)[1]
#mean_bcnn_w_v = mean_bcnn_w .* ones(bcnn_s)
#x_bcnn = collect(1:bcnn_s)
fcnn_w = removewatt(fcnn[:, 1])
mem_fcnn = removewatt(fcnn[:, 3])
#val_ch = fcnn_w .> 28
#fcnn_start = findall(x -> x==1, val_ch)[1]
#fcnn_end = findall(x -> x==1, val_ch)[end]
#fcnn_w_tot = fcnn_w[fcnn_start:fcnn_end]
#fcnn_tot = sum(fcnn_w_tot)
#mean_fcnn_w = mean(fcnn_w_tot)
#fcnn_s = size(fcnn_w_tot)[1]
#mean_fcnn_w_v = mean_fcnn_w .* ones(fcnn_s)
#x_fcnn = collect(1:fcnn_s)
mem_bcnn = parse.(Int64, replace.(bcnn[:, 3], r"[^0-9]+" => ""))
mem_fcnn = parse.(Int64, replace.(fcnn[:, 3], r"[^0-9]+" => ""))
#en_plot = boxplot(bcnn_w_tot,x_bcnn,title="Energy per sample",xlabel="Watts",ylabel="No of samples",label="BCNN",seriesalpha=0.5)
#boxplot!(fcnn_w_tot,x_fcnn,label="CNN",seriesalpha=0.5,color="green")
#savefig(en_plot,"energy.png")
#en_plot = histogram(bcnn_w,xlabel="Watts",ylabel="No of samples",label="BCNN",color="blue",normalize=:pdf,bar_position = :overlay)
#histogram!(fcnn_w,label="LeNet",seriesalpha=0.9,color="#ff9900",normalize=:pdf,bar_position=:overlay)
#savefig(en_plot,"energy.png")
#bcnn_w = incwzeros(bcnn_w,length(fcnn_w)-length(bcnn_w))
#bcnn_w = reshape(bcnn_w,:,1)
#fcnn_w = reshape(fcnn_w,:,1)
#w_mat = hcat(bcnn_w,fcnn_w)
#groupedbar(rand(length(bcnn_w),2),bar_width=0.5,bar_position = :dodge)
#w_mat = reshape(w_mat,2,:)
#mem_plot = plot(mem_fcnn,title="Memory usage, one epoch in MiB's",label="CNN Memory",lw=2,seriesalpha=0.5)
#plot!(mem_bcnn,label="BCNN Memory",lw=2,seriesalpha=0.5,fmt=:png)
#savefig(mem_plot,"memory.png")
#mem_plot = histogram(mem_bcnn,xlabel="MiB",ylabel="No of samples",label="BCNN",color="blue")
#histogram!(mem_fcnn,label="LeNet",seriesalpha=0.9,color="#ff9900")
#savefig(mem_plot,"memory.png")
#plot(bcnn_w_tot,title="BCNN energy in one epoch",label="Watt",lw=2);plot!(mean_bcnn_w_v,label="Mean")
#plot(fcnn_w_tot,title="CNN energy in one epoch",label="Watt",lw=2);plot!(mean_fcnn_w_v,label="Mean")
#scatter(bcnn[1,:],title="Energy usage in W",label="Bayesian CNN",lw=2);scatter!(fcnn[1,1:463],label="CNN",lw=2)
bcnn_d = countmap(bcnn_w)
bcnn_u = unique(bcnn_w)
bcnn_v = Vector()
for u in bcnn_u
push!(bcnn_v, bcnn_d[u])
end
m_bcnn_d = countmap(mem_bcnn)
m_bcnn_u = unique(mem_bcnn)
m_bcnn_v = Vector()
for u in m_bcnn_u
push!(m_bcnn_v, m_bcnn_d[u])
end
fcnn_d = countmap(fcnn_w)
fcnn_u = unique(fcnn_w)
fcnn_v = Vector()
for u in fcnn_u
push!(fcnn_v, fcnn_d[u])
end
m_fcnn_d = countmap(mem_fcnn)
m_fcnn_u = unique(mem_fcnn)
m_fcnn_v = Vector()
for u in fcnn_u
push!(m_fcnn_v, fcnn_d[u])
end
bcnn_v = bcnn_v / norm(bcnn_v)
fcnn_v = fcnn_v / norm(fcnn_v)
m_bcnn_v = m_bcnn_v / norm(m_bcnn_v)
m_fcnn_v = m_fcnn_v / norm(m_fcnn_v)
en_plot = plot(
[
bar(x = bcnn_u, y = bcnn_v, name = "BCNN", marker_color = "blue"),
bar(x = fcnn_u, y = fcnn_v, name = "LeNet", marker_color = "#ff9900"),
],
Layout(
barmode = "group",
xaxis_tickangle = -45,
yaxis_title_text = "Samples",
xaxis_title_text = "Watts",
title = "Energy distribution experiment 1",
),
)
savefig(en_plot, "energy.png")
mem_plot = plot(
[
bar(x = m_bcnn_u, y = m_bcnn_v, name = "BCNN", marker_color = "blue"),
bar(x = m_fcnn_u, y = m_fcnn_v, name = "LeNet", marker_color = "#ff9900"),
],
Layout(
barmode = "group",
xaxis_tickangle = -45,
yaxis_title_text = "Samples",
xaxis_title_text = "MiB",
title = "MRW distribution experiment 1",
),
)
savefig(mem_plot, "memory.png")

1563
process_m_data.jl Executable file
View File

File diff suppressed because it is too large Load Diff

1272
process_w_data.jl Executable file
View File

File diff suppressed because it is too large Load Diff

295
sum_w_data.jl Executable file
View File

@ -0,0 +1,295 @@
include("aux_func.jl")
using LinearAlgebra
using Statistics
using PlotlyJS
#=
Obtain energy data
=#
#mni_folder_1 = "ini_exp_data/"
mni_folder_100 = "exp_100_epochs/";
cif_folder_100 = "CIFAR_100_epoch/";
mni_folder_acc = "data_bounded/";
cif_folder_acc = "CIFAR_acc_bound/";
mni_folder_wat = "data_budget/";
cif_folder_wat = "CIFAR_energy_bound/";
mni_folder_est = "early_stop_res/";
cif_folder_est = "CIFAR_early_stop/";
bayes_model = "bayes";
freq_model = "freq";
w_type = "watt";
e_type = "exp";
#=
Load GPU data
=#
mni_100_bay_ene = getgpudata(mni_folder_100, bayes_model, w_type, "eo");
mni_100_fre_ene = getgpudata(mni_folder_100, freq_model, w_type, "eo");
cif_100_bay_ene = getgpudata(cif_folder_100, bayes_model, w_type, "eo");
cif_100_fre_ene = getgpudata(cif_folder_100, freq_model, w_type, "eo");
mni_acc_bay_ene = getgpudata(mni_folder_acc, bayes_model, w_type, "eo");
mni_acc_fre_ene = getgpudata(mni_folder_acc, freq_model, w_type, "eo");
cif_acc_bay_ene = getgpudata(cif_folder_acc, bayes_model, w_type, "eo");
cif_acc_fre_ene = getgpudata(cif_folder_acc, freq_model, w_type, "eo");
mni_wat_bay_ene = getgpudata(mni_folder_wat, bayes_model, w_type, "eo");
mni_wat_fre_ene = getgpudata(mni_folder_wat, freq_model, w_type, "eo");
cif_wat_bay_ene = getgpudata(cif_folder_wat, bayes_model, w_type, "eo");
cif_wat_fre_ene = getgpudata(cif_folder_wat, freq_model, w_type, "eo");
mni_est_bay_ene = getgpudata(mni_folder_est, bayes_model, w_type, "eo");
mni_est_fre_ene = getgpudata(mni_folder_est, freq_model, w_type, "eo");
cif_est_bay_ene = getgpudata(cif_folder_est, bayes_model, w_type, "eo");
cif_est_fre_ene = getgpudata(cif_folder_est, freq_model, w_type, "eo");
#=
Define CPU paths
=#
cif = "cifar";
mni = "mnist";
bay = "bayes";
wat = "cpu_watts";
frq = "freq";
ram = "ram_use";
_100 = "100";
acc = "acc";
es = "es";
wbud = "wbud";
#=
Load CPU data
=#
bay_cif_100_ene = readcpudata(cif, bay, wat, _100);
bay_mni_100_ene = readcpudata(mni, bay, wat, _100);
frq_cif_100_ene = readcpudata(cif, frq, wat, _100);
frq_mni_100_ene = readcpudata(mni, frq, wat, _100);
bay_cif_acc_ene = readcpudata(cif, bay, wat, acc);
bay_mni_acc_ene = readcpudata(mni, bay, wat, acc);
frq_cif_acc_ene = readcpudata(cif, frq, wat, acc);
frq_mni_acc_ene = readcpudata(mni, frq, wat, acc);
bay_cif_wbu_ene = readcpudata(cif, bay, wat, wbud);
bay_mni_wbu_ene = readcpudata(mni, bay, wat, wbud);
frq_cif_wbu_ene = readcpudata(cif, frq, wat, wbud);
frq_mni_wbu_ene = readcpudata(mni, frq, wat, wbud);
bay_mni_est_ene = readcpudata(mni, bay, wat, es);
bay_cif_est_ene = readcpudata(cif, bay, wat, es);
frq_cif_est_ene = readcpudata(cif, frq, wat, es);
frq_mni_est_ene = readcpudata(mni, frq, wat, es);
for s = 1:5
bay_cif_acc_ene[s] = round.(getcpuwatt(bay_cif_acc_ene[s]))
bay_cif_est_ene[s] = round.(getcpuwatt(bay_cif_est_ene[s]))
bay_cif_wbu_ene[s] = round.(getcpuwatt(bay_cif_wbu_ene[s]))
bay_cif_100_ene[s] = round.(getcpuwatt(bay_cif_100_ene[s]))
bay_mni_acc_ene[s] = round.(getcpuwatt(bay_mni_acc_ene[s]))
bay_mni_est_ene[s] = round.(getcpuwatt(bay_mni_est_ene[s]))
bay_mni_wbu_ene[s] = round.(getcpuwatt(bay_mni_wbu_ene[s]))
bay_mni_100_ene[s] = round.(getcpuwatt(bay_mni_100_ene[s]))
frq_cif_acc_ene[s] = round.(getcpuwatt(frq_cif_acc_ene[s]))
frq_cif_est_ene[s] = round.(getcpuwatt(frq_cif_est_ene[s]))
frq_cif_wbu_ene[s] = round.(getcpuwatt(frq_cif_wbu_ene[s]))
frq_cif_100_ene[s] = round.(getcpuwatt(frq_cif_100_ene[s]))
frq_mni_acc_ene[s] = round.(getcpuwatt(frq_mni_acc_ene[s]))
frq_mni_est_ene[s] = round.(getcpuwatt(frq_mni_est_ene[s]))
frq_mni_wbu_ene[s] = round.(getcpuwatt(frq_mni_wbu_ene[s]))
frq_mni_100_ene[s] = round.(getcpuwatt(frq_mni_100_ene[s]))
end
for s = 1:5
mni_100_bay_ene[s]["Ene"] = vcat(mni_100_bay_ene[s]["Ene"], bay_mni_100_ene[s])
mni_100_fre_ene[s]["Ene"] = vcat(mni_100_fre_ene[s]["Ene"], frq_mni_100_ene[s])
cif_100_bay_ene[s]["Ene"] = vcat(cif_100_bay_ene[s]["Ene"], bay_cif_100_ene[s])
cif_100_fre_ene[s]["Ene"] = vcat(cif_100_fre_ene[s]["Ene"], frq_cif_100_ene[s])
mni_acc_bay_ene[s]["Ene"] = vcat(mni_acc_bay_ene[s]["Ene"], bay_mni_acc_ene[s])
mni_acc_fre_ene[s]["Ene"] = vcat(mni_acc_fre_ene[s]["Ene"], frq_mni_acc_ene[s])
cif_acc_bay_ene[s]["Ene"] = vcat(cif_acc_bay_ene[s]["Ene"], bay_cif_acc_ene[s])
cif_acc_fre_ene[s]["Ene"] = vcat(cif_acc_fre_ene[s]["Ene"], frq_cif_acc_ene[s])
mni_wat_bay_ene[s]["Ene"] = vcat(mni_wat_bay_ene[s]["Ene"], bay_mni_wbu_ene[s])
mni_wat_fre_ene[s]["Ene"] = vcat(mni_wat_fre_ene[s]["Ene"], frq_mni_wbu_ene[s])
cif_wat_bay_ene[s]["Ene"] = vcat(cif_wat_bay_ene[s]["Ene"], bay_cif_wbu_ene[s])
cif_wat_fre_ene[s]["Ene"] = vcat(cif_wat_fre_ene[s]["Ene"], frq_cif_wbu_ene[s])
mni_est_bay_ene[s]["Ene"] = vcat(mni_est_bay_ene[s]["Ene"], bay_mni_est_ene[s])
mni_est_fre_ene[s]["Ene"] = vcat(mni_est_fre_ene[s]["Ene"], frq_mni_est_ene[s])
cif_est_bay_ene[s]["Ene"] = vcat(cif_est_bay_ene[s]["Ene"], bay_cif_est_ene[s])
cif_est_fre_ene[s]["Ene"] = vcat(cif_est_fre_ene[s]["Ene"], frq_cif_est_ene[s])
end
all_data_ene = Dict(
"mni" => Dict(
1 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
2 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
3 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
4 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
5 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
),
"cif" => Dict(
1 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
2 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
3 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
4 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
5 => Dict(
"100" => Dict("bay" => 0.0, "frq" => 0.0),
"acc" => Dict("bay" => 0.0, "frq" => 0.0),
"wat" => Dict("bay" => 0.0, "frq" => 0.0),
"est" => Dict("bay" => 0.0, "frq" => 0.0),
),
),
)
for s = 1:5
all_data_ene["mni"][s]["100"]["bay"] = sum(mni_100_bay_ene[s]["Ene"])
all_data_ene["mni"][s]["100"]["frq"] = sum(mni_100_fre_ene[s]["Ene"])
all_data_ene["cif"][s]["100"]["bay"] = sum(cif_100_bay_ene[s]["Ene"])
all_data_ene["cif"][s]["100"]["frq"] = sum(cif_100_fre_ene[s]["Ene"])
all_data_ene["mni"][s]["acc"]["bay"] = sum(mni_acc_bay_ene[s]["Ene"])
all_data_ene["mni"][s]["acc"]["frq"] = sum(mni_acc_fre_ene[s]["Ene"])
all_data_ene["cif"][s]["acc"]["bay"] = sum(cif_acc_bay_ene[s]["Ene"])
all_data_ene["cif"][s]["acc"]["frq"] = sum(cif_acc_fre_ene[s]["Ene"])
all_data_ene["mni"][s]["wat"]["bay"] = sum(mni_wat_bay_ene[s]["Ene"])
all_data_ene["mni"][s]["wat"]["frq"] = sum(mni_wat_fre_ene[s]["Ene"])
all_data_ene["cif"][s]["wat"]["bay"] = sum(cif_wat_bay_ene[s]["Ene"])
all_data_ene["cif"][s]["wat"]["frq"] = sum(cif_wat_fre_ene[s]["Ene"])
all_data_ene["mni"][s]["est"]["bay"] = sum(mni_est_bay_ene[s]["Ene"])
all_data_ene["mni"][s]["est"]["frq"] = sum(mni_est_fre_ene[s]["Ene"])
all_data_ene["cif"][s]["est"]["bay"] = sum(cif_est_bay_ene[s]["Ene"])
all_data_ene["cif"][s]["est"]["frq"] = sum(cif_est_fre_ene[s]["Ene"])
end
#=
# sums
=#
println(
"MNIST LeNet 100, Total Energy: [$(all_data_ene["mni"][1]["100"]["frq"]),$(all_data_ene["mni"][2]["100"]["frq"]),$(all_data_ene["mni"][3]["100"]["frq"]),$(all_data_ene["mni"][4]["100"]["frq"]),$(all_data_ene["mni"][5]["100"]["frq"])]",
)
println(
"MNIST Bayes 100, Total Energy: [$(all_data_ene["mni"][1]["100"]["bay"]),$(all_data_ene["mni"][2]["100"]["bay"]),$(all_data_ene["mni"][3]["100"]["bay"]),$(all_data_ene["mni"][4]["100"]["bay"]),$(all_data_ene["mni"][5]["100"]["bay"])]",
)
println(
"CIFAR LeNet 100, Total Energy: [$(all_data_ene["cif"][1]["100"]["frq"]),$(all_data_ene["cif"][2]["100"]["frq"]),$(all_data_ene["cif"][3]["100"]["frq"]),$(all_data_ene["cif"][4]["100"]["frq"]),$(all_data_ene["cif"][5]["100"]["frq"])]",
)
println(
"CIFAR Bayes 100, Total Energy: [$(all_data_ene["cif"][1]["100"]["bay"]),$(all_data_ene["cif"][2]["100"]["bay"]),$(all_data_ene["cif"][3]["100"]["bay"]),$(all_data_ene["cif"][4]["100"]["bay"]),$(all_data_ene["cif"][5]["100"]["bay"])]",
)
println(
"MNIST LeNet Acc, Total Energy: [$(all_data_ene["mni"][1]["acc"]["frq"]),$(all_data_ene["cif"][2]["acc"]["frq"]),$(all_data_ene["cif"][3]["acc"]["frq"]),$(all_data_ene["cif"][4]["100"]["frq"]),$(all_data_ene["cif"][5]["acc"]["frq"])]",
)
println(
"MNIST Bayes Acc, Total Energy: [$(all_data_ene["mni"][1]["acc"]["bay"]),$(all_data_ene["cif"][2]["acc"]["bay"]),$(all_data_ene["cif"][3]["acc"]["bay"]),$(all_data_ene["cif"][4]["100"]["bay"]),$(all_data_ene["cif"][5]["acc"]["bay"])]",
)
println(
"CIFAR LeNet Acc, Total Energy: [$(all_data_ene["cif"][1]["acc"]["frq"]),$(all_data_ene["cif"][2]["acc"]["frq"]),$(all_data_ene["cif"][3]["acc"]["frq"]),$(all_data_ene["cif"][4]["100"]["frq"]),$(all_data_ene["cif"][5]["acc"]["frq"])]",
)
println(
"CIFAR Bayes Acc, Total Energy: [$(all_data_ene["cif"][1]["acc"]["bay"]),$(all_data_ene["cif"][2]["acc"]["bay"]),$(all_data_ene["cif"][3]["acc"]["bay"]),$(all_data_ene["cif"][4]["100"]["bay"]),$(all_data_ene["cif"][5]["acc"]["bay"])]",
)
println(
"MNIST LeNet Wat, Total Energy: [$(all_data_ene["mni"][1]["wat"]["frq"]),$(all_data_ene["cif"][2]["wat"]["frq"]),$(all_data_ene["cif"][3]["wat"]["frq"]),$(all_data_ene["cif"][4]["wat"]["frq"]),$(all_data_ene["cif"][5]["wat"]["frq"])]",
)
println(
"MNIST Bayes Wat, Total Energy: [$(all_data_ene["mni"][1]["wat"]["bay"]),$(all_data_ene["cif"][2]["wat"]["bay"]),$(all_data_ene["cif"][3]["wat"]["bay"]),$(all_data_ene["cif"][4]["wat"]["bay"]),$(all_data_ene["cif"][5]["wat"]["bay"])]",
)
println(
"CIFAR LeNet Wat, Total Energy: [$(all_data_ene["cif"][1]["wat"]["frq"]),$(all_data_ene["cif"][2]["wat"]["frq"]),$(all_data_ene["cif"][3]["wat"]["frq"]),$(all_data_ene["cif"][4]["wat"]["frq"]),$(all_data_ene["cif"][5]["wat"]["frq"])]",
)
println(
"CIFAR Bayes Wat, Total Energy: [$(all_data_ene["cif"][1]["wat"]["bay"]),$(all_data_ene["cif"][2]["wat"]["bay"]),$(all_data_ene["cif"][3]["wat"]["bay"]),$(all_data_ene["cif"][4]["wat"]["bay"]),$(all_data_ene["cif"][5]["wat"]["bay"])]",
)
println(
"MNIST LeNet Est, Total Energy: [$(all_data_ene["mni"][1]["est"]["frq"]),$(all_data_ene["cif"][2]["est"]["frq"]),$(all_data_ene["cif"][3]["est"]["frq"]),$(all_data_ene["cif"][4]["est"]["frq"]),$(all_data_ene["cif"][5]["est"]["frq"])]",
)
println(
"MNIST Bayes Est, Total Energy: [$(all_data_ene["mni"][1]["est"]["bay"]),$(all_data_ene["cif"][2]["est"]["bay"]),$(all_data_ene["cif"][3]["est"]["bay"]),$(all_data_ene["cif"][4]["est"]["bay"]),$(all_data_ene["cif"][5]["est"]["bay"])]",
)
println(
"CIFAR LeNet Est, Total Energy: [$(all_data_ene["cif"][1]["est"]["frq"]),$(all_data_ene["cif"][2]["est"]["frq"]),$(all_data_ene["cif"][3]["est"]["frq"]),$(all_data_ene["cif"][4]["est"]["frq"]),$(all_data_ene["cif"][5]["est"]["frq"])]",
)
println(
"CIFAR Bayes Est, Total Energy: [$(all_data_ene["cif"][1]["est"]["bay"]),$(all_data_ene["cif"][2]["est"]["bay"]),$(all_data_ene["cif"][3]["est"]["bay"]),$(all_data_ene["cif"][4]["est"]["bay"]),$(all_data_ene["cif"][5]["est"]["bay"])]",
)

127
test.jl Executable file
View File

@ -0,0 +1,127 @@
24546
8675
14405
8251
12414
1054
16014
1730
3411
3309
3125
3416
26037
10727
25228
9074
32151
1870
17650
1990
4006
3518
4843
3705
33038
8497
19262
8289
16115
858
14456
1153
3264
3007
3148
3155
43922
9162
28756
9055
34825
1523
15070
1268
6794
3293
4594
3670
41450
8579
24245
8244
20672
599
15082
1171
3352
2178
3184
2887
50435
10065
32195
8995
37124
1103
15984
1391
6232
2394
3445
3215
49003
8751
28977
8364
27099
595
20822
944
3353
1872
3229
2250
55590
9538
34952
9258
41751
888
22713
1045
4642
2064
4118
3153
55256
8843
33954
8442
32833
433
22218
967
3073
1727
3092
1907
59022
10711
36849
10154
45188
503
23276
2735
3954
2413
3632
3211

17
times.jl Executable file
View File

@ -0,0 +1,17 @@
using Dates
using Dates: Time
include("aux_func.jl")
freq_results = getimes("times_frequentist");
bayes_results = getimes("times_bayesian");
freq_results = concat(freq_results);
bayes_results = concat(bayes_results);
io = open("result_times_freq", "w");
write(io, freq_results);
close(io);
io = open("result_times_bayes", "w");
write(io, bayes_results);
close(io);