This repository has been archived by the owner on Nov 17, 2023. It is now read-only.
Improve sparse adagrad update #9651
Merged
Conversation
This file contains 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
ZiyueHuang
reviewed
Feb 1, 2018
src/operator/optimizer_op-inl.h
Outdated
| } | ||
| const DType grad_squared = grad_rescaled * grad_rescaled; | ||
| state_data[data_j] += grad_squared; | ||
| const DType div = grad_rescaled / math::sqrt(state_data[data_j] + epsilon); |
There was a problem hiding this comment.
nit : use mshadow_op::square_root::Map? Seems that other kernels use mshadow_op instead of directly call math func
src/operator/optimizer_op-inl.h
Outdated
| state_data[data_j] += grad_squared; | ||
| const DType div = grad_rescaled / math::sqrt(state_data[data_j] + epsilon); | ||
| // No need to use KERNEL_ASSIGN, as we already checked req is kWriteInplace | ||
| out_data[data_j] = weight_data[data_j] + div * -lr; |
ZiyueHuang
approved these changes
Feb 1, 2018
cjolivier01
approved these changes
Feb 1, 2018
szha
reviewed
Feb 1, 2018
| 'rescale_grad': self.rescale_grad} | ||
| if self.clip_gradient: | ||
| kwargs['clip_gradient'] = self.clip_gradient | ||
| sparse.adagrad_update(weight, grad, history, out=weight, lr=lr, wd=wd, **kwargs) |
There was a problem hiding this comment.
since wd is still passed to the operator, we can just check whether wd is 0 or not in the backend.
szha
reviewed
Feb 1, 2018
| float wd; | ||
| DMLC_DECLARE_PARAMETER(AdagradParam) { | ||
| DMLC_DECLARE_FIELD(lr) | ||
| .describe("Learning rate"); |
There was a problem hiding this comment.
It's indented. What's the concern?
|
Is this ready to go in? |
eric-haibin-lin
changed the title
|
Hi everyone, Considering the |
eric-haibin-lin
added a commit
to eric-haibin-lin/mxnet
that referenced
this pull request
Mar 4, 2018
* fix adagrad * add test * fix lint * CR comments * remove raise in python * enhance unit test * revert wd changes * revert dense op changes
jinhuang415
pushed a commit
to jinhuang415/incubator-mxnet
that referenced
this pull request
Mar 30, 2018
* fix adagrad * add test * fix lint * CR comments * remove raise in python * enhance unit test * revert wd changes * revert dense op changes
rahul003
pushed a commit
to rahul003/mxnet
that referenced
this pull request
Jun 4, 2018
* fix adagrad * add test * fix lint * CR comments * remove raise in python * enhance unit test * revert wd changes * revert dense op changes
zheng-da
pushed a commit
to zheng-da/incubator-mxnet
that referenced
this pull request
Jun 28, 2018
* fix adagrad * add test * fix lint * CR comments * remove raise in python * enhance unit test * revert wd changes * revert dense op changes
7 tasks
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
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.
Description
RowSparseNDArray.indicesandRowSparseNDArray.data.@ZiyueHuang @piiswrong @sxjscience @cjolivier01 @szha
Commit author is mistaken. Nvm.
Checklist
Essentials
make lint)Changes
Comments