Skip to content
New issue

Have a question about this project? # for a free GitHub account to open an issue and contact its maintainers and the community.

#

By clicking “#”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? # to your account

Add a dropout_mask method for ComplexF64 array types #1572

Closed
wants to merge 3 commits into from
Closed

Add a dropout_mask method for ComplexF64 array types #1572

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
f09e3fe
added complex array method for dropout_mask
ShoofLLC Apr 10, 2021
8133af3
Update normalisation.jl
ShoofLLC Apr 10, 2021
db69b3a
Update normalisation.jl
ShoofLLC Apr 10, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 7 additions & 1 deletion src/layers/normalise.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,6 +46,12 @@ function dropout_mask(x, p; dims=:)
return y
end

function dropout_mask(x::Array{Complex{Float64}}, p; dims=:)
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Seems a bit strict. Maybe we want to generalise the existing method? It should work the same for single precision.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I agree, I just didn't want to introduce changes that would affect any other types

Copy link
Contributor Author

@ShoofLLC ShoofLLC Apr 10, 2021 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I added a test with a complex array as an input, I could also just modify dropout_mask to always use a Float type when calling similar:

Modifying the dropout_mask function itself to always have a float32 type when calling similar:
function dropout_mask(x, p; dims=:) y = rand!(similar(x, Float32, _dropout_shape(x, dims))) y .= _dropout_kernel.(y, p, 1 - p) return y end

would make dropout work for any input type, currently using dropout for Int inputs would fail with an inexact error, but I didn't want to modify the main method just for this use-case.

This might actually save some memory because if a network uses double-precision the rand would be double as well, which is not really needed just for comparison with p.

y = rand!(similar(x, Float64, _dropout_shape(x, dims)))
y .= _dropout_kernel.(y, p, 1 - p)
return y
end

"""
Dropout(p; dims=:)

Expand Down Expand Up @@ -457,4 +463,4 @@ scale parameters, `false` otherwise.

See [`BatchNorm`](@ref), [`InstanceNorm`](@ref), [`GroupNorm`](@ref), and [`LayerNorm`](@ref).
"""
hasaffine(l::Union{BatchNorm, InstanceNorm, LayerNorm, GroupNorm}) = l.affine
hasaffine(l::Union{BatchNorm, InstanceNorm, LayerNorm, GroupNorm}) = l.affine
5 changes: 5 additions & 0 deletions test/layers/normalisation.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,11 @@ using Zygote: pullback
evalwgrad(f, x...) = pullback(f, x...)[1]

@testset "Dropout" begin
x = [1.0+0im,2.0+1im,3.0+3im]
@test x == Dropout(0.1)(x)
@test x == evalwgrad(Dropout(0), x)
@test zero(x) == evalwgrad(Dropout(1), x)

x = [1.,2.,3.]
@test x == Dropout(0.1)(x)
@test x == evalwgrad(Dropout(0), x)
Expand Down