This repository has been archived by the owner on Nov 17, 2023. It is now read-only.
Mark test_dropout as flaky #19553
Merged
Mark test_dropout as flaky #19553
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
Two issues. Issue 1: #14288 Issue 2: [2020-11-17T06:58:34.678Z] def check_passthrough(ratio, shape, cudnn_off=True): [2020-11-17T06:58:34.678Z] # test inference_mode forward and then backward [2020-11-17T06:58:34.678Z] a = mx.random.uniform(shape=shape) [2020-11-17T06:58:34.678Z] a.attach_grad() [2020-11-17T06:58:34.678Z] with mx.autograd.record(train_mode=False): [2020-11-17T06:58:34.678Z] b = mx.nd.Dropout(a, ratio, cudnn_off=cudnn_off) # dropout acts as identity [2020-11-17T06:58:34.678Z] b.backward() [2020-11-17T06:58:34.678Z] assert_almost_equal(a.grad.asnumpy(), mx.nd.ones_like(b).asnumpy()) [2020-11-17T06:58:34.678Z] [2020-11-17T06:58:34.678Z] shape = (100, 100) [2020-11-17T06:58:34.678Z] check_dropout_ratio(0.5, shape) [2020-11-17T06:58:34.678Z] check_dropout_ratio(0.0, shape) [2020-11-17T06:58:34.678Z] > check_dropout_ratio(1.0, shape) [...] [2020-11-17T06:58:34.678Z] # Hopefully should be within ratio/2 % [2020-11-17T06:58:34.678Z] error = abs(output_sum - input_sum) / input_sum [2020-11-17T06:58:34.678Z] if ratio == 1.0: [2020-11-17T06:58:34.678Z] > assert output_zeroes == len(input) [2020-11-17T06:58:34.678Z] E assert 9999 == 10000 [2020-11-17T06:58:34.678Z] E +9999 [2020-11-17T06:58:34.678Z] E -10000
|
Hey @leezu , Thanks for submitting the PR
CI supported jobs: [centos-cpu, clang, edge, unix-cpu, windows-cpu, windows-gpu, centos-gpu, miscellaneous, unix-gpu, sanity, website] Note: |
lanking520
added
the
pr-awaiting-testing
szha
approved these changes
Nov 17, 2020
lanking520
added
pr-work-in-progress
|
@mxnet-bot run ci [windows-gpu] |
|
Jenkins CI successfully triggered : [windows-gpu] |
lanking520
added
pr-awaiting-testing
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.
Two issues.
Issue 1: #14288
Issue 2:
[2020-11-17T06:58:34.678Z] def check_passthrough(ratio, shape, cudnn_off=True):
[2020-11-17T06:58:34.678Z] # test inference_mode forward and then backward
[2020-11-17T06:58:34.678Z] a = mx.random.uniform(shape=shape)
[2020-11-17T06:58:34.678Z] a.attach_grad()
[2020-11-17T06:58:34.678Z] with mx.autograd.record(train_mode=False):
[2020-11-17T06:58:34.678Z] b = mx.nd.Dropout(a, ratio, cudnn_off=cudnn_off) # dropout acts as identity
[2020-11-17T06:58:34.678Z] b.backward()
[2020-11-17T06:58:34.678Z] assert_almost_equal(a.grad.asnumpy(), mx.nd.ones_like(b).asnumpy())
[2020-11-17T06:58:34.678Z]
[2020-11-17T06:58:34.678Z] shape = (100, 100)
[2020-11-17T06:58:34.678Z] check_dropout_ratio(0.5, shape)
[2020-11-17T06:58:34.678Z] check_dropout_ratio(0.0, shape)
[2020-11-17T06:58:34.678Z] > check_dropout_ratio(1.0, shape)
[...]
[2020-11-17T06:58:34.678Z] # Hopefully should be within ratio/2 %
[2020-11-17T06:58:34.678Z] error = abs(output_sum - input_sum) / input_sum
[2020-11-17T06:58:34.678Z] if ratio == 1.0:
[2020-11-17T06:58:34.678Z] > assert output_zeroes == len(input)
[2020-11-17T06:58:34.678Z] E assert 9999 == 10000
[2020-11-17T06:58:34.678Z] E +9999
[2020-11-17T06:58:34.678Z] E -10000