TastyPancakes
/
Flux.jl
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Updated Msle loss

This commit is contained in:
Adarsh Kumar 2020-02-27 16:26:17 +05:30 committed by GitHub
parent 3d8965230f
commit 9dce623214
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 5 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
src/layers/stateless.jl
View File

@ -17,33 +17,28 @@ mse(ŷ, y) = sum((ŷ .- y).^2) * 1 // length(y)
""" """
mean_squared_logarithmic_error(ŷ, y;ϵ1=eps.(Float64.(ŷ)),ϵ2=eps.(Float64.(y))) msle(ŷ, y;ϵ1=eps.(Float64.(ŷ)),ϵ2=eps.(Float64.(y)))
L2 loss function. Returns the mean of the squared logarithmic errors of prediction ŷ, and true values y. The ϵ1 and ϵ2 terms provide numerical stability. Mean Squared Logarithmic Error,an L2 loss function. Returns the mean of the squared logarithmic errors of prediction ŷ, and true values y. The ϵ1 and ϵ2 terms provide numerical stability.
(Computes mean of squared(log(predicted values)-log(true value)). This error penalizes an under-predicted estimate greater than an over-predicted estimate. (Computes mean of squared(log(predicted values)-log(true value)). This error penalizes an under-predicted estimate greater than an over-predicted estimate.
```julia ```julia
julia> y_=[14726,327378,74734] julia> y=[14726,327378,74734]
3-element Array{Int64,1}: 3-element Array{Int64,1}:
14726 14726
327378 327378
74734 74734
julia> y = [12466.1,16353.95,16367.98] julia> ŷ = [12466.1,16353.95,16367.98]
3-element Array{Float64,1}: 3-element Array{Float64,1}:
12466.1 12466.1
16353.95 16353.95
16367.98 16367.98
julia> mean_squared_logarithmic_error(y,y_) julia> msle(ŷ,y)
3.771271382334686 3.771271382334686
``` ```
Alias:
msle(ŷ,y;ϵ1=eps.(Float64.(ŷ)),ϵ2=eps.(Float64.(y)))
""" """
mean_squared_logarithmic_error(ŷ, y;ϵ1=eps.(ŷ),ϵ2=eps.(eltype(ŷ).(y))) = sum((log.(ŷ+ϵ1).-log.(y+ϵ2)).^2) * 1 // length(y)
#Alias
msle(ŷ, y;ϵ1=eps.(ŷ),ϵ2=eps.(eltype(ŷ).(y))) = sum((log.(ŷ+ϵ1).-log.(y+ϵ2)).^2) * 1 // length(y) msle(ŷ, y;ϵ1=eps.(ŷ),ϵ2=eps.(eltype(ŷ).(y))) = sum((log.(ŷ+ϵ1).-log.(y+ϵ2)).^2) * 1 // length(y)