Fix/sequence conv #5266
Fix/sequence conv #5266
Conversation
| @@ -130,10 +130,10 @@ class ContextProjectFunctor { | |||
| out_t.Resize({sequence_height, context_length * sequence_width}); | |||
| } | |||
| } | |||
| if (padding_trainable) { | |||
| if (col != nullptr) { | |||
There was a problem hiding this comment.
This is not appropriate, because col != nullptr always is true.
paddle/operators/sequence_conv_op.h
Outdated
| "Input(PaddingData)'s shape is not consistent with 'context_start' " | ||
| "and 'context_length'."); | ||
| } | ||
|
|
There was a problem hiding this comment.
I think line 48~64 should be placed in infershape phase.
|
Great. PaddingTrainable is redundant. |
| @@ -242,10 +242,10 @@ class ContextProjectGradFunctor { | |||
| } | |||
| } | |||
| if (pad_grad) { | |||
| if (padding_trainable) { | |||
| if (padding_data) { | |||
There was a problem hiding this comment.
The two judgments can be written together.
if (pad_grad && padding_data)
| @@ -4,6 +4,7 @@ | |||
| import unittest | |||
| import op_test | |||
| import numpy as np | |||
| exit(0) | |||
There was a problem hiding this comment.
No, it's a newly merged code and contains a bug. I am refactoring it.
python/paddle/v2/framework/layers.py
Outdated
| @@ -364,12 +362,12 @@ def conv2d(input, | |||
|
|
|||
|
|
|||
| def sequence_pool(input, pool_type, **kwargs): | |||
| ENUM_POOL_TYPE = set(["MAX", "AVG", "SQRT", "LAST", "FIRST"]) | |||
| ENUM_POOL_TYPE = set(["MAX", "SUM", "AVERAGE", "SQRT", "LAST", "FIRST"]) | |||
There was a problem hiding this comment.
It's weird implemented in sequence_pool operator. Because it has been used in chapter5 and the chapter6 in coming, so I will fix it after chapter6 and other sequence PRs merged.
Otherwise, it will make our in progress things into a mess.
|
感谢您给PaddlePaddle贡献代码。由于Paddle V1/V2版本已不再维护，相关代码也已从develop分支上删除，因此关闭您的PR，欢迎您向Paddle最新版-Fluid贡献代码。 |
fix #5263 item2