Fix MSELoss when target.requires_grad is True. #44437
Closed
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
MSELoss had a completely different (and incorrect, see #43228) path when target.requires_grad was True. This PR does the following: 1) adds derivative support for target via the normal derivatives.yaml route 2) kill the different (and incorrect) path for when target.requires_grad was True 3) modify the MSELoss CriterionTests to verify that the target derivative is checked. TODO: 1) do we still need check_criterion_jacobian when we run grad/gradgrad checks? 2) ensure the Module tests check when target.requires_grad 3) do we actually test when reduction='none' and reduction='mean'? [ghstack-poisoned]
This was referenced Sep 9, 2020
gchanan
added a commit
that referenced
this pull request
Sep 9, 2020
MSELoss had a completely different (and incorrect, see #43228) path when target.requires_grad was True. This PR does the following: 1) adds derivative support for target via the normal derivatives.yaml route 2) kill the different (and incorrect) path for when target.requires_grad was True 3) modify the MSELoss CriterionTests to verify that the target derivative is checked. TODO: 1) do we still need check_criterion_jacobian when we run grad/gradgrad checks? 2) ensure the Module tests check when target.requires_grad 3) do we actually test when reduction='none' and reduction='mean'? ghstack-source-id: cbfba5f Pull Request resolved: #44437
💊 CI failures summary and remediationsAs of commit c472bb2 (more details on the Dr. CI page):
1 failure confirmed as flaky and can be ignored:
ci.pytorch.org: 1 failedThis comment was automatically generated by Dr. CI (expand for details).Follow this link to opt-out of these comments for your Pull Requests.Please report bugs/suggestions on the GitHub issue tracker or post in the (internal) Dr. CI Users group. This comment has been revised 8 times. |
Contributor
Author
|
|
Codecov Report
@@ Coverage Diff @@
## gh/gchanan/322/base #44437 +/- ##
=======================================================
- Coverage 68.00% 67.99% -0.01%
=======================================================
Files 382 382
Lines 49379 49373 -6
=======================================================
- Hits 33578 33572 -6
Misses 15801 15801
Continue to review full report at Codecov.
|
MSELoss had a completely different (and incorrect, see #43228) path when target.requires_grad was True. This PR does the following: 1) adds derivative support for target via the normal derivatives.yaml route 2) kill the different (and incorrect) path for when target.requires_grad was True 3) modify the MSELoss CriterionTests to verify that the target derivative is checked. TODO: 1) do we still need check_criterion_jacobian when we run grad/gradgrad checks? 2) ensure the Module tests check when target.requires_grad 3) do we actually test when reduction='none' and reduction='mean'? Differential Revision: [D23612166](https://our.internmc.facebook.com/intern/diff/D23612166) [ghstack-poisoned]
gchanan
requested review from
ebetica,
goldsborough and
yf225
as code owners
gchanan
added a commit
that referenced
this pull request
Sep 10, 2020
MSELoss had a completely different (and incorrect, see #43228) path when target.requires_grad was True. This PR does the following: 1) adds derivative support for target via the normal derivatives.yaml route 2) kill the different (and incorrect) path for when target.requires_grad was True 3) modify the MSELoss CriterionTests to verify that the target derivative is checked. TODO: 1) do we still need check_criterion_jacobian when we run grad/gradgrad checks? 2) ensure the Module tests check when target.requires_grad 3) do we actually test when reduction='none' and reduction='mean'? ghstack-source-id: 88dab44 Pull Request resolved: #44437
albanD
approved these changes
Sep 10, 2020
Collaborator
albanD
left a comment
albanD
left a comment
There was a problem hiding this comment.
lgtm
Not sure why we didn't do that before... :D
albanD
reviewed
Sep 10, 2020
| # currently compute the gradient w.r.t. target for loss functions. | ||
| gradcheck(apply_fn, inputs) | ||
| if target.requires_grad: | ||
| inputs = inputs + (target,) |
Collaborator
There was a problem hiding this comment.
I think this works but it is a bit fragile.
Your apply_fn above actually ignores the target that is given in *params. But it works because it captures the same target from the parent scope. And the gradcheck ensures that you can do that. But I think it would be clearer to just get the target from the apply_fn inputs properly no?
albanD
approved these changes
Sep 10, 2020
Contributor
xuzhao9
pushed a commit
that referenced
this pull request
Sep 18, 2020
Summary: Pull Request resolved: #44437 MSELoss had a completely different (and incorrect, see #43228) path when target.requires_grad was True. This PR does the following: 1) adds derivative support for target via the normal derivatives.yaml route 2) kill the different (and incorrect) path for when target.requires_grad was True 3) modify the MSELoss CriterionTests to verify that the target derivative is checked. TODO: 1) do we still need check_criterion_jacobian when we run grad/gradgrad checks? 2) ensure the Module tests check when target.requires_grad 3) do we actually test when reduction='none' and reduction='mean'? Test Plan: Imported from OSS Reviewed By: albanD Differential Revision: D23612166 Pulled By: gchanan fbshipit-source-id: 4f74d38d8a81063c74e002e07fbb7837b2172a10
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.
Stack from ghstack:
MSELoss had a completely different (and incorrect, see #43228) path when target.requires_grad was True.
This PR does the following:
TODO:
Differential Revision: D23612166