[AutoDiff] Add Differentiable.zeroTangentVectorInitializer.
#28416
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
The `Differentiable.zeroTangentVectorInitializer` requirement will enable correct, efficient reverse-mode differentiation of struct property accesses. A `zeroTangentVectorInitializer` closure is more efficient than a `zeroTangentVector` computed property, which would always capture `self`. Add `Differentiable.zeroTangentVector` default instance property, which returns `self.zeroTangentVectorInitializer()`. Todos: - Implement derived conformances for `zeroTangentVectorInitializer`. - Implement differentiation transform support for `zeroTangentVectorInitializer` and struct/tuple projection instructions.
|
I didn't add any tests for now because I don't think there's anything meaningful to test yet. |
marcrasi
approved these changes
Nov 21, 2019
|
@swift-ci Please test tensorflow |
Also for forward-mode, because JVP should return a differential that does not capture unnecessary information from the original code.
My earlier work in progress can be found here: rxwei@936791e. |
rxwei
approved these changes
Nov 21, 2019
|
@swift-ci Please test tensorflow linux |
|
I'll merge after filing issues for todos ( Edit: TF-1005 tracks per-instance zero tangent vector initialization. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The
Differentiable.zeroTangentVectorInitializerrequirement will enablecorrect, efficient differentiation of struct property and tuple element accesses.
A
zeroTangentVectorInitializerclosure is more efficient than azeroTangentVectorcomputed property, which would always captureself.Add
Differentiable.zeroTangentVectordefault instance property, whichreturns
self.zeroTangentVectorInitializer().Todos:
zeroTangentVectorInitializer.zeroTangentVectorInitializerand struct/tuple projection instructions.Mirror of #28310 (with minor edits), which was merged but became lost in the
tensorflowbranch commit history for some reason.