Add SELU activation function #1769

fmassa · 2017-06-10T19:45:47Z

Also fixes a problem with RReLU in inplace mode.

apaszke

All contbuilds are failing

torch/nn/_functions/thnn/activation.py

+    def __init__(self, inplace=False):
+        super(SELU, self).__init__(inplace)
+        self.alpha = 1.6732632423543772848170429916717
+        self.scale = 1.0507009873554804934193349852946


torch/nn/modules/activation.py

+    More details can be found in the paper `Self-Normalizing Neural Networks`_ .
+
+    Args:
+        inplace: can optionally do the operation in-place


fmassa · 2017-06-10T20:13:14Z

Yeah, I just saw that the tests are failing because there is no legacy equivalent for SELU. Should we add it or just skip on SELU?

apaszke · 2017-06-10T20:13:57Z

No, you only have to move the test spec from common_nn.py to test_nn.py

Remove in-place tests for RReLU. For some reason they fail on legacy nn, but passes on nn

It also supports double backprop, verifyed with gradgradcheck

fmassa · 2017-06-10T23:24:19Z

Added new-style Function to SELU. Double backprop was verified with gradgradcheck.

torch/nn/_functions/thnn/activation.py


+class SELU(InplaceFunction):
+    alpha = 1.6732632423543772848170429916717
+    scale = 1.0507009873554804934193349852946


…08e7e3 Summary: Previous import was dc75285d4a1cff9618400164dfdb26c5a1bab70a Included changes: - **[15c33c9](onnx/onnx@15c33c9)**: Add ppc64le build (pytorch#1768) <Chin Huang> - **[198f840](onnx/onnx@198f840)**: Update Broadcasting.md (pytorch#1769) <Verma-Rajat> - **[60ac95f](onnx/onnx@60ac95f)**: Merge back from release 1.4.1 (pytorch#1767) <Raymond Yang> - **[a683372](onnx/onnx@a683372)**: Bump up version number for v1.4.0 (pytorch#1761) (pytorch#1763) <Raymond Yang> - **[dbf3581](onnx/onnx@dbf3581)**: Add TfIdfVectorizer operator to ONNX (pytorch#1721) <Dmitri Smirnov> Differential Revision: D13858840 fbshipit-source-id: 90b2e21c80de4936507a27fc93d0879128ab4fb7

…08e7e3 (#16493) Summary: Pull Request resolved: #16493 Previous import was dc75285d4a1cff9618400164dfdb26c5a1bab70a Included changes: - **[15c33c9](onnx/onnx@15c33c9)**: Add ppc64le build (#1768) <Chin Huang> - **[198f840](onnx/onnx@198f840)**: Update Broadcasting.md (#1769) <Verma-Rajat> - **[60ac95f](onnx/onnx@60ac95f)**: Merge back from release 1.4.1 (#1767) <Raymond Yang> - **[a683372](onnx/onnx@a683372)**: Bump up version number for v1.4.0 (#1761) (#1763) <Raymond Yang> - **[dbf3581](onnx/onnx@dbf3581)**: Add TfIdfVectorizer operator to ONNX (#1721) <Dmitri Smirnov> Reviewed By: zrphercule Differential Revision: D13858840 fbshipit-source-id: 1d00f63f265cc6deed965b92ed00c44f547ff03e

This refactors `TransformPropagator` as a 3-level abstraction: `TransformPropagator` is a subclass of `MaxRootDomainInfoPropagator`, while `MaxRootDomainInfoPropagator` is a subclass of `MaxInfoPropagator`. `MaxInfoPropagator` implements the Dijkstra algorithm for propagating on the DAG, but it has no knowledge what "information" we are trying to preserve and what we are propagating. `MaxRootDomainInfoPropagator` inherits `MaxInfoPropagator` for preserving most root/rfactor domain information. But it does not have any knowledge about what we are propagating. `TransformPropagator` further inherits `MaxRootDomainInfoPropagator` and propagates transformations of leaf IDs.

Syncing nvfuser devel branch to upstream master. https://github.com/csarofeen/pytorch/ Code changes includes: - TransformPropagator refactor: switched to Dijkstra instead of exhaustive enumeration on all possible paths to reduce compilation time on transform propagation; - Indexing refactor: remove reference tensor creation in all tensor indexing logic (#1690) - (more) generic grouped grid reduction kernel; - Minor parser/fuser patches: 1. zero-dim tensor reduction support 3. no-op binary removal within fused graph 4. expand supported in fusion Squashed commits to WAR github API Commits that's actually in this PR from the devel branch: ``` a054b3e Refactor TransormPropagator to allow specifying a position and propagating to part of the DAG (#1775) d67e1cd Indexing refactor stage 1: remove reference tensor creation in all tensor indexing logic (#1690) 1b65299 Issue 1770 (#1774) 35b0427 Avoid compilation errors like below: (#1773) 452c773 Ignore reductions of zero-dim tensors per PyTorch conventions (#1771) 31d6c56 TransformPropagator refactor (#1769) 570c5a8 Merge pull request #1767 from csarofeen/upstream_merge_0621 9d6c3d8 merging upstream 61305cd 0ed815f New TransformPropagator algorithm (#1763) 6c19520 no-op binary removal (#1764) ec7fa41 Proper propagation of IterType (#1762) b263562 Fix dimensionality check (#1759) 2d6343f More generic grouped grid reduction kernel (#1740) 64e2b56 [nvfuser] prevent spamming warning message (#77777) (#1758) 0c43162 [nvFuser] Improving bitwise ops support (#77158) (#1757) b93a147 Parser expand (#1754) ``` RUN_TORCHBENCH: nvfuser Pull Request resolved: #80355 Approved by: https://github.com/davidberard98

Summary: Syncing nvfuser devel branch to upstream master. https://github.com/csarofeen/pytorch/ Code changes includes: - TransformPropagator refactor: switched to Dijkstra instead of exhaustive enumeration on all possible paths to reduce compilation time on transform propagation; - Indexing refactor: remove reference tensor creation in all tensor indexing logic (#1690) - (more) generic grouped grid reduction kernel; - Minor parser/fuser patches: 1. zero-dim tensor reduction support 3. no-op binary removal within fused graph 4. expand supported in fusion Squashed commits to WAR github API Commits that's actually in this PR from the devel branch: ``` a054b3e Refactor TransormPropagator to allow specifying a position and propagating to part of the DAG (#1775) d67e1cd Indexing refactor stage 1: remove reference tensor creation in all tensor indexing logic (#1690) 1b65299 Issue 1770 (#1774) 35b0427 Avoid compilation errors like below: (#1773) 452c773 Ignore reductions of zero-dim tensors per PyTorch conventions (#1771) 31d6c56 TransformPropagator refactor (#1769) 570c5a8 Merge pull request #1767 from csarofeen/upstream_merge_0621 9d6c3d8 merging upstream 61305cd 0ed815f New TransformPropagator algorithm (#1763) 6c19520 no-op binary removal (#1764) ec7fa41 Proper propagation of IterType (#1762) b263562 Fix dimensionality check (#1759) 2d6343f More generic grouped grid reduction kernel (#1740) 64e2b56 [nvfuser] prevent spamming warning message (#77777) (#1758) 0c43162 [nvFuser] Improving bitwise ops support (#77158) (#1757) b93a147 Parser expand (#1754) ``` RUN_TORCHBENCH: nvfuser Pull Request resolved: #80355 Reviewed By: qihqi Differential Revision: D37573400 Pulled By: davidberard98 fbshipit-source-id: 52ab68d89ec01ef61f69f5abeb18c9d3a312aa64

fmassa added 4 commits June 10, 2017 18:52

Add SELU activation function

aa32667

Remove unnecessary case

d94c419

Add Function for SELU + tests and fix RReLU inplace

79753ab

Fix extra line in doc

3ca2317

apaszke reviewed Jun 10, 2017

View reviewed changes

fmassa added 2 commits June 10, 2017 21:47

Fix tests

ef9b7c1

Remove in-place tests for RReLU. For some reason they fail on legacy nn, but passes on nn

SELU in new-style Function

bd41e81

It also supports double backprop, verifyed with gradgradcheck

apaszke approved these changes Jun 10, 2017

View reviewed changes

Fix flake8

07a386c

soumith merged commit a24db91 into pytorch:master Jun 11, 2017

fmassa deleted the selu branch June 11, 2017 08:21

andreh7 mentioned this pull request Jul 25, 2017

Add Alpha Dropout #1775

Merged

andreh7 reviewed Jul 25, 2017

View reviewed changes

torch/nn/_functions/thnn/activation.py

class SELU(InplaceFunction):

alpha = 1.6732632423543772848170429916717

scale = 1.0507009873554804934193349852946

This comment was marked as off-topic.

Sign in to view

houseroad mentioned this pull request Jan 29, 2019

Automatic update of fbcode/onnx to 15c33c945851907411619f599900c3852108e7e3 #16493

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Add SELU activation function #1769

Add SELU activation function #1769

Uh oh!

fmassa commented Jun 10, 2017

Uh oh!

apaszke left a comment

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

fmassa commented Jun 10, 2017

Uh oh!

apaszke commented Jun 10, 2017

Uh oh!

fmassa commented Jun 10, 2017

Uh oh!

This comment was marked as off-topic.

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Add SELU activation function #1769

Add SELU activation function #1769

Uh oh!

Conversation

fmassa commented Jun 10, 2017

Uh oh!

apaszke left a comment

Choose a reason for hiding this comment

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

fmassa commented Jun 10, 2017

Uh oh!

apaszke commented Jun 10, 2017

Uh oh!

fmassa commented Jun 10, 2017

Uh oh!

This comment was marked as off-topic.

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants