Skip to content
New issue

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

Sign up for GitHub

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

Already on GitHub? Sign in to your account

support for dims in sum(f, x; dims) #684

Merged
merged 1 commit into from
Jun 16, 2020
Merged

support for dims in sum(f, x; dims) #684

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
11 changes: 8 additions & 3 deletions src/lib/array.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -223,9 +223,14 @@ end
end
end

function _pullback(cx::AContext, ::typeof(sum), f, xs::AbstractArray)
y, back = pullback(cx, ((f, xs) -> sum(f.(xs))), f, xs)
y, ȳ -> (nothing, back(ȳ)...)
_normalize_kws(kws::NamedTuple) = kws
_normalize_kws(kws) = NamedTuple()

function _pullback(cx::AContext, kwtype, kws, ::typeof(sum), f, xs::AbstractArray)
norm_kws = _normalize_kws(kws)
@assert !haskey(norm_kws, :init) # TODO add init support (julia 1.6)
y, back = pullback(cx, (f, xs) -> sum(f.(xs); norm_kws...), f, xs)
y, ȳ -> (nothing, nothing, nothing, back(ȳ)...)
end

@adjoint function sum(::typeof(abs2), X::AbstractArray; dims = :)
Expand Down
50 changes: 27 additions & 23 deletions test/gradcheck.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -103,29 +103,33 @@ end
@test gradtest((w, x) -> parent(w)*x, randn(5,5)', randn(5,5))
@test gradtest((w, x) -> parent(w)*x, transpose(randn(5,5)), randn(5,5))

@test gradtest(x -> sum(x, dims = (2, 3)), (3,4,5))
@test gradtest(x -> sum(abs2, x), randn(4, 3, 2))
@test gradtest(x -> sum(abs2, x; dims=1), randn(4, 3, 2))
@test gradtest(x -> sum(x[i] for i in 1:length(x)), randn(10))
@test gradtest(x -> sum(i->x[i], 1:length(x)), randn(10)) # https://github.com/FluxML/Zygote.jl/issues/231
@test gradtest(x -> sum((i->x[i]).(1:length(x))), randn(10))

# https://github.com/FluxML/Zygote.jl/issues/314
@test gradient((x,y) -> sum(yi -> yi*x, y), 1, [1,1]) == (2, [1, 1])
@test gradient((x,y) -> prod(yi -> yi*x, y), 1, [1,1]) == (2, [1, 1])

@test gradient((x,y) -> sum(map(yi -> yi*x, y)), 1, [1,1]) == (2, [1, 1])
@test gradient((x,y) -> prod(map(yi -> yi*x, y)), 1, [1,1]) == (2, [1, 1])

@test gradtest(x -> prod(x, dims = (2, 3)), (3,4,5))
@test gradtest(x -> prod(x), (3,4))
@test gradient(x -> prod(x), (1,2,3))[1] == (6,3,2)

@test gradtest(x -> cumsum(x, dims=2), (3,4,5))
@test gradtest(x -> cumsum(x, dims=1), (3,))
@test gradtest(x -> cumsum(x), (4,))
@test gradtest(x -> cumsum(x, dims=3), (5,)) # trivial
@test gradtest(x -> cumsum(x, dims=3), (3,4)) # trivial
@testset "sum, prod, cumsum" begin
@test gradtest(x -> sum(x, dims = (2, 3)), (3,4,5))
@test gradtest(x -> sum(abs2, x), randn(4, 3, 2))
@test gradtest(x -> sum(abs2, x; dims=1), randn(4, 3, 2))
@test gradtest(x -> sum(x[i] for i in 1:length(x)), randn(10))
@test gradtest(x -> sum(i->x[i], 1:length(x)), randn(10)) # issue #231
@test gradtest(x -> sum((i->x[i]).(1:length(x))), randn(10))
@test gradtest(X -> sum(x -> x^2, X), randn(10))
@test gradtest(X -> sum(sum(x -> x^2, X; dims=1)), randn(10)) # issue #681

# https://github.com/FluxML/Zygote.jl/issues/314
@test gradient((x,y) -> sum(yi -> yi*x, y), 1, [1,1]) == (2, [1, 1])
@test gradient((x,y) -> prod(yi -> yi*x, y), 1, [1,1]) == (2, [1, 1])

@test gradient((x,y) -> sum(map(yi -> yi*x, y)), 1, [1,1]) == (2, [1, 1])
@test gradient((x,y) -> prod(map(yi -> yi*x, y)), 1, [1,1]) == (2, [1, 1])

@test gradtest(x -> prod(x, dims = (2, 3)), (3,4,5))
@test gradtest(x -> prod(x), (3,4))
@test gradient(x -> prod(x), (1,2,3))[1] == (6,3,2)

@test gradtest(x -> cumsum(x, dims=2), (3,4,5))
@test gradtest(x -> cumsum(x, dims=1), (3,))
@test gradtest(x -> cumsum(x), (4,))
@test gradtest(x -> cumsum(x, dims=3), (5,)) # trivial
@test gradtest(x -> cumsum(x, dims=3), (3,4)) # trivial
end

@test gradtest(x -> softmax(x).*(1:3), 3)
@test gradtest(x -> softmax(x).*(1:3), (3,5))
Expand Down