Create performance tips docs section (#615)

* Create performance_tips.jl

* Rename performance_tips.jl to performance_tips.md

* add perf tips

* Update docs/src/performance_tips.md

Co-Authored-By: oxinabox <oxinabox@ucc.asn.au>

* Update docs/src/performance_tips.md

Co-Authored-By: oxinabox <oxinabox@ucc.asn.au>

* Update make.jl

* Update and rename performance_tips.md to performance.md

* spelling

* Update docs/src/performance.md

Co-Authored-By: oxinabox <oxinabox@ucc.asn.au>

* Update docs/src/performance.md

Co-Authored-By: oxinabox <oxinabox@ucc.asn.au>

* Update performance.md

* Update performance.md

* Update docs/src/performance.md

Co-Authored-By: oxinabox <oxinabox@ucc.asn.au>

* Update docs/src/performance.md

Co-Authored-By: oxinabox <oxinabox@ucc.asn.au>
This commit is contained in:
Lyndon White 2019-02-19 15:03:41 +00:00 committed by Mike J Innes
parent 78876a14b3
commit ebf50f4e1c
2 changed files with 77 additions and 0 deletions

View File

@ -19,6 +19,7 @@ makedocs(modules=[Flux, NNlib],
"One-Hot Encoding" => "data/onehot.md",
"GPU Support" => "gpu.md",
"Saving & Loading" => "saving.md",
"Performance Tips" => "performance.md",
"Internals" =>
["Backpropagation" => "internals/tracker.md"],
"Community" => "community.md"])

76
docs/src/performance.md Normal file
View File

@ -0,0 +1,76 @@
# Performance Tips
All the usual [Julia performance tips apply](https://docs.julialang.org/en/v1/manual/performance-tips/).
As always [profiling your code](https://docs.julialang.org/en/v1/manual/profile/#Profiling-1) is generally a useful way of finding bottlenecks.
Below follow some Flux specific tips/reminders.
## Don't use more precision than you need.
Flux works great with all kinds of number types.
But often you do not need to be working with say `Float64` (let alone `BigFloat`).
Switching to `Float32` can give you a significant speed up,
not because the operations are faster, but because the memory usage is halved.
Which means allocations occur much faster.
And you use less memory.
## Make sure your custom activation functions preserve the type of their inputs
Not only should your activation functions be [type-stable](https://docs.julialang.org/en/v1/manual/performance-tips/#Write-%22type-stable%22-functions-1),
they should also preserve the type of their inputs.
A very artificial example using an activatioon function like
```
my_tanh(x) = Float64(tanh(x))
```
will result in performance on `Float32` input orders of magnitude slower than the normal `tanh` would,
because it results in having to use slow mixed type multiplication in the dense layers.
Which means if you change your data say from `Float64` to `Float32` (which should give a speedup: see above),
you will see a large slow-down
This can occur sneakily, because you can cause type-promotion by interacting with a numeric literals.
E.g. the following will have run into the same problem as above:
```
leaky_tanh(x) = 0.01x + tanh(x)
```
While one could change your activation function (e.g. to use `0.01f0x`) to avoid this when ever your inputs change,
the idiomatic (and safe way) is to use `oftype`.
```
leaky_tanh(x) = oftype(x/1, 0.01) + tanh(x)
```
## Evaluate batches as Matrices of features, rather than sequences of Vector features
While it can sometimes be tempting to process your observations (feature vectors) one at a time
e.g.
```julia
function loss_total(xs::AbstractVector{<:Vector}, ys::AbstractVector{<:Vector})
sum(zip(xs, ys)) do (x, y_target)
y_pred = model(x) # evaluate the model
return loss(y_pred, y_target)
end
end
```
It is much faster to concatenate them into a matrix,
as this will hit BLAS matrix-matrix multiplication, which is much faster than the equivalent sequence of matrix-vector multiplications.
Even though this means allocating new memory to store them contiguously.
```julia
x_batch = reduce(hcat, xs)
y_batch = reduce(hcat, ys)
...
function loss_total(x_batch::Matrix, y_batch::Matrix)
y_preds = model(x_batch)
sum(loss.(y_preds, y_batch))
end
```
When doing this kind of concatenation use `reduce(hcat, xs)` rather than `hcat(xs...)`.
This will avoid the splatting penality, and will hit the optimised `reduce` method.