Skip to content

Conversation

@JiayiFeng
Copy link
Collaborator

fixes #4406

@JiayiFeng JiayiFeng requested review from pkuyym and reyoung September 27, 2017 00:26
return (1. - val) * (1. - val)
else:
return 0
return 0.
Copy link
Collaborator

Choose a reason for hiding this comment

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

wow


def test_check_grad(self):
self.check_grad(['X'], 'Out', max_relative_error=0.005)
self.check_grad(['X'], 'Out', max_relative_error=0.01)
Copy link
Collaborator

Choose a reason for hiding this comment

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

Just a random question: could we have a systematic way to set the error tolerance?

Copy link
Collaborator

Choose a reason for hiding this comment

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

There are a lot of indifferentiable points for forwarding operators. Only writers of operators could know which point is indifferentiable.

So I think that should be managed manually by Op writers.

Copy link
Collaborator

@reyoung reyoung left a comment

Choose a reason for hiding this comment

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

Excellent.

@reyoung reyoung merged commit 7b385ff into PaddlePaddle:develop Sep 27, 2017
@JiayiFeng JiayiFeng deleted the fix_huber_loss_test_error branch September 27, 2017 02:41
luotao1 pushed a commit that referenced this pull request Nov 9, 2022
* fix paddle.get_default_dtype 

Chinese and English return values are inconsistent

* fix paddle.matmul 文档评估 #4407

把函数的输出改成正确的

* fix paddle.std文档评估 #4370

增加了一个unbiased=False的代码示例,没有增加numpy,怕引起误会。

* fix paddle.load文档测评 #4455 

只把代码拆分了5段

* try

* try

* try

* Update io.py

* Update io.py

* Update creation.py

* Update creation.py

* [Docs]add name description

* [Docs]fix broadcasting issue

Co-authored-by: Ligoml <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Error of computational accuracy in the unit test of Modified Huber Loss Op

3 participants