This repository has been archived by the owner on Nov 17, 2023. It is now read-only.
Support sparse for custom python operators #8620
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
python/mxnet/operator.py
Outdated
| return in_stype, [in_stype[0]]*len(self.list_outputs()), \ | ||
| [in_stype[0]]*len(self.list_auxiliary_states()) | ||
|
|
||
| def infer_storage_type(self, in_stype): |
There was a problem hiding this comment.
nit: would it be more natural to put def infer_storage_type before def infer_storage_type_backward?
python/mxnet/operator.py
Outdated
| Returns | ||
| ------- | ||
| in_stype : list | ||
| list of argument stypes. Can be modified from in_stype. |
There was a problem hiding this comment.
What does Can be modified from in_stype mean? Can they be modified when it's already set?
There was a problem hiding this comment.
No it cannot be modified if it is already set. Can be modified only if it is undefined storage type. I have removed the line to avoid confusion
src/operator/custom/custom.cc
Outdated
| for (size_t i = 0; i < oattr->size(); i++) { | ||
| STORAGE_TYPE_ASSIGN_CHECK(*oattr, i, kDefaultStorage); | ||
| } | ||
| dispatch_mode_assign(dispatch_mode, DispatchMode::kFComputeEx); |
There was a problem hiding this comment.
Should we use DISPATCH_MODE_ASSIGN_CHECK instead?
| if in_data[0].stype == 'default': | ||
| aux[0][:] = 1 | ||
| self.assign(out_data[0], req[0], in_data[0]*in_data[0]) | ||
| else: |
There was a problem hiding this comment.
could you add check here:
if data.stype == 'csr':
assert(isinstance(data, CSRNDArray))
| check_numeric_gradient(op, [x], [aux]) | ||
|
|
||
| x = mx.nd.array(np.random.uniform(-1, 1, size=(4, 10))) |
There was a problem hiding this comment.
Is this line duplicated on purpose?
| @@ -3570,12 +3570,19 @@ def test_rcbrt_op(): | |||
| def test_custom_op(): | |||
| class Sqr(mx.operator.CustomOp): | |||
| def forward(self, is_train, req, in_data, out_data, aux): | |||
| self.assign(out_data[0], req[0], in_data[0]*in_data[0]) | |||
| aux[0][:] = 1 | |||
| #self.assign(out_data[0], req[0], in_data[0]*in_data[0]) | |||
eric-haibin-lin
approved these changes
Nov 16, 2017
piiswrong
added a commit
to piiswrong/mxnet
that referenced
this pull request
Nov 20, 2017
This reverts commit 938eda9.
cjolivier01
pushed a commit
that referenced
this pull request
Nov 21, 2017
eric-haibin-lin
pushed a commit
to eric-haibin-lin/mxnet
that referenced
this pull request
Dec 3, 2017
* Add asscipy support and coo format support * Comment misalignment change * Add documentation for Sparse NDarray * Change comment * Adding comments and support for dtype * Modifying tests * Add spsp None check * Fix lint * Custom operators for sparse * Use DISPATCH_MODE_ASSIGN_CHECK * Change NDArray to _ndarray_cls * Remove redundant code * Add a test to make sure the NDArray is an instance of CSRNDArray * Fix lint * Fix test * Trigger CI
eric-haibin-lin
pushed a commit
to eric-haibin-lin/mxnet
that referenced
this pull request
Dec 3, 2017
…ache#8733) This reverts commit 938eda9.
zhreshold
pushed a commit
to zhreshold/mxnet
that referenced
this pull request
Dec 14, 2017
…ache#8733) This reverts commit 938eda9.
rahul003
pushed a commit
to rahul003/mxnet
that referenced
this pull request
Jun 4, 2018
* Add asscipy support and coo format support * Comment misalignment change * Add documentation for Sparse NDarray * Change comment * Adding comments and support for dtype * Modifying tests * Add spsp None check * Fix lint * Custom operators for sparse * Use DISPATCH_MODE_ASSIGN_CHECK * Change NDArray to _ndarray_cls * Remove redundant code * Add a test to make sure the NDArray is an instance of CSRNDArray * Fix lint * Fix test * Trigger CI
rahul003
pushed a commit
to rahul003/mxnet
that referenced
this pull request
Jun 4, 2018
…ache#8733) This reverts commit 938eda9.
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
This PR is to support sparse for custom python operators. Specifically, one should be able to use sparse storage types with custom python operators, right now only default storage types are supported.
Checklist
Essentials
make lint)Changes
Comments
@piiswrong @eric-haibin-lin