Skip to content

[AutoDiff] [API] Fix bug in Tensor.reshaped VJP #21923

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

Merged
merged 1 commit into from
Jan 17, 2019
Merged

[AutoDiff] [API] Fix bug in Tensor.reshaped VJP #21923

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
2 changes: 1 addition & 1 deletion stdlib/public/TensorFlow/Gradients.swift
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -518,7 +518,7 @@ extension Tensor where Scalar : Differentiable & FloatingPoint {
) -> (Tensor, (Tensor) -> Tensor) {
let value = reshaped(toShape: newShape)
return (value, { v in
return v.reshaped(toShape: newShape)
return v.reshaped(toShape: self.shapeTensor)
Copy link
Contributor

Choose a reason for hiding this comment

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

Nit:

  • We generally don't use self. unless it's required.
  • It'd be more efficient if you make it capture self.shapeTensor explicitly instead of capturing self. I'd write this as
{ [shape = shapeTensor] v in
  return v.reshaped(toShape: shape)
}

Copy link
Contributor

@dan-zheng dan-zheng Jan 16, 2019 •
edited
Loading

Choose a reason for hiding this comment

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

FYI: there are definitely more places where we could use explicit captures instead of capturing self for efficiency. We should change those places at some point!

Copy link
Contributor

Choose a reason for hiding this comment

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

Depends on whether self is practically a larger tensor that the result you get from operations on self :)

})
}

Expand Down
12 changes: 10 additions & 2 deletions test/TensorFlowRuntime/tensor_autodiff_runtime.swift
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -92,16 +92,24 @@ TensorADTests.testAllBackends("mean") {
expectTrue(meanGradAlongAxes(input) == expected)
}

TensorADTests.testAllBackends("reshaped") {
let shapeTensor = Tensor<Int32>([2, 2, 2])
let input = Tensor<Float>(ones: [2, 4])
let reshapedPullback = pullback(at: input) { (a: Tensor<Float>) in a.reshaped(toShape: shapeTensor) }
let reshaped = Tensor<Float>(ones: [2, 2, 2])
expectTrue(reshapedPullback(reshaped) == input)
}

TensorADTests.testAllBackends("transposed") {
let input = Tensor<Float>(ones: [2, 3])
let transposed = Tensor<Float>(ones: [3, 2])
let transposedPullback = pullback(at: input) { (a: Tensor<Float>) in a.transposed() }
let transposedPermutationsPullback = pullback(at: input) { (a: Tensor<Float>) in a.transposed(withPermutations: [1, 0]) }
let transposedVariadiicsPullback = pullback(at: input) { (a: Tensor<Float>) in a.transposed(withPermutations: 1, 0) }
let transposedVariadicsPullback = pullback(at: input) { (a: Tensor<Float>) in a.transposed(withPermutations: 1, 0) }

expectTrue(transposedPullback(transposed) == input)
expectTrue(transposedPermutationsPullback(transposed) == input)
expectTrue(transposedVariadiicsPullback(transposed) == input)
expectTrue(transposedVariadicsPullback(transposed) == input)
}

TensorADTests.testAllBackends("relu") {
Expand Down