Skip to content

Fix pointwise loss broadcast #12996

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

Jump to bottom
Closed
ailzhang wants to merge 11 commits into from
Closed

Fix pointwise loss broadcast #12996

ailzhang wants to merge 11 commits into from

Conversation

@ailzhang
Copy link
Contributor

@ailzhang ailzhang commented Oct 23, 2018

Fixes #12129 , #12327
cc: @fmassa _pointwise_loss is currently only called by l1_loss, mse_loss and smooth_l1_loss which all makes sense to have broadcast.

facebook-github-bot
facebook-github-bot reviewed Oct 23, 2018
View reviewed changes
Copy link
Contributor

@facebook-github-bot facebook-github-bot left a comment

Choose a reason for hiding this comment

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

ailzhang has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

colesbury
colesbury reviewed Oct 23, 2018
View reviewed changes
torch/nn/functional.py
reduction = _Reduction.get_enum(reduction)
return torch._C._nn.smooth_l1_loss(input, target, reduction)
reduction = _Reduction.legacy_get_string(size_average, reduce)
return _pointwise_loss(

This comment was marked as off-topic.

Sign in to view

This comment was marked as off-topic.

Sign in to view

This comment was marked as off-topic.

Sign in to view

@ailzhang
Copy link
Contributor Author

ailzhang commented Oct 24, 2018

@pytorchbot retest this please

@ssnl ssnl closed this Oct 24, 2018
@ssnl ssnl reopened this Oct 24, 2018
fmassa
fmassa reviewed Oct 24, 2018
View reviewed changes
torch/nn/functional.py Outdated
return torch._C._nn.smooth_l1_loss(input, target, reduction)
reduction = _Reduction.legacy_get_string(size_average, reduce)
return _pointwise_loss(
lambda a, b: torch.where(torch.abs(a - b) < 1, 0.5 * (torch.abs(a - b)) ** 2, torch.abs(a - b) - 0.5),

This comment was marked as off-topic.

Sign in to view

facebook-github-bot
facebook-github-bot reviewed Oct 25, 2018
View reviewed changes
Copy link
Contributor

@facebook-github-bot facebook-github-bot left a comment

Choose a reason for hiding this comment

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

ailzhang has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

@ssnl
Copy link
Collaborator

ssnl commented Oct 25, 2018

can we have a test?

facebook-github-bot
facebook-github-bot reviewed Oct 25, 2018
View reviewed changes
Copy link
Contributor

@facebook-github-bot facebook-github-bot left a comment

Choose a reason for hiding this comment

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

ailzhang has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

ssnl
ssnl reviewed Oct 25, 2018
View reviewed changes
test/test_nn.py Outdated
# When target.requires_grad=True, its impl is in Python, while the other is in TH.
target = torch.ones(2, 10, requires_grad=requires_grad)
l = fn(input, target, 'none')
self.assertEqual(l.size(), target.size())

This comment was marked as off-topic.

Sign in to view

@ailzhang ailzhang closed this Oct 26, 2018
@ailzhang ailzhang reopened this Oct 26, 2018
facebook-github-bot
facebook-github-bot reviewed Oct 26, 2018
View reviewed changes
Copy link
Contributor

@facebook-github-bot facebook-github-bot left a comment

Choose a reason for hiding this comment

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

ailzhang has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

facebook-github-bot
facebook-github-bot reviewed Oct 29, 2018
View reviewed changes
Copy link
Contributor

@facebook-github-bot facebook-github-bot left a comment

Choose a reason for hiding this comment

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

ailzhang has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

ssnl
ssnl reviewed Oct 30, 2018
View reviewed changes
test/test_nn.py Outdated
for requires_grad in (True, False):
# When target.requires_grad=True, its impl is in Python, while the other is in TH.
target = torch.randn(2, 10, requires_grad=requires_grad)
l = fn(input, target, 'none')

This comment was marked as off-topic.

Sign in to view

facebook-github-bot
facebook-github-bot reviewed Oct 30, 2018
View reviewed changes
Copy link
Contributor

@facebook-github-bot facebook-github-bot left a comment

Choose a reason for hiding this comment

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

ailzhang has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

facebook-github-bot pushed a commit that referenced this pull request Oct 31, 2018
@facebook-github-bot
Fix pointwise loss broadcast (#12996)
488d393 
Summary: Fixes #12129 , #12327

Differential Revision: D10513781

Pulled By: ailzhang

fbshipit-source-id: a210008a39ff6c3f056c9fbe3f0576cfcce638ec
@ailzhang
Copy link
Contributor Author

ailzhang commented Oct 31, 2018

This has been landed to master, not sure why it didn't get closed tho. Closing manually.

@ailzhang ailzhang closed this Oct 31, 2018
@ailzhang ailzhang mentioned this pull request Oct 31, 2018
Closed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@colesbury colesbury colesbury left review comments

@facebook-github-bot facebook-github-bot facebook-github-bot left review comments

@fmassa fmassa fmassa left review comments

@ssnl ssnl ssnl approved these changes

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

5 participants

@ailzhang @ssnl @colesbury @facebook-github-bot @fmassa