fix bugs of parsing network topology in text generation task #2384
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
|
Do not approve. |
luotao1
reviewed
Jun 6, 2017
doc/api/v2/config/layer.rst
Outdated
| Miscs | ||
| ===== | ||
|
|
||
| dropout_layer |
There was a problem hiding this comment.
v2 api中默认后面不加layer，dropout_layer后面为什么要加呢
There was a problem hiding this comment.
已经修改。将 drop 从 network 下移动到 layer 下，潜在会影响一些旧的配置。
lcy-seso
force-pushed
the
fix_config_parsing_bug
branch
2 times, most recently
from
f6a2934
to
0e16aed
Compare
lcy-seso
force-pushed
the
fix_config_parsing_bug
branch
from
0e16aed
to
9146e7b
Compare
lcy-seso
changed the title
lcy-seso
force-pushed
the
fix_config_parsing_bug
branch
from
9146e7b
to
862b72b
Compare
lcy-seso
force-pushed
the
fix_config_parsing_bug
branch
from
862b72b
to
692d251
Compare
|
Please ignore "TeamCity build failed", see: #2401 |
|
@helinwang I get it. Thank you. |
lcy-seso
commented
Jun 7, 2017
| @@ -235,11 +235,10 @@ def __trim_submodel__(old_submodel, layer_names, input_layer_names, | |||
| def parse_network(output_layers, extra_layers=None): | |||
| if not isinstance(output_layers, collections.Sequence): | |||
| output_layers = [output_layers] | |||
| if extra_layers is not None and not isinstance(extra_layers, | |||
| collections.Sequence): | |||
There was a problem hiding this comment.
- Topology 的构造函数中会对 extra_layers 的类型进行检查
- 检查的过程中，如果传入的extra_layer 不是 python list，会转化为 list
https://github.com/PaddlePaddle/Paddle/blob/develop/python/paddle/v2/topology.py#L43 - Topology 构造函数调用
parse_network走到这里时，如果指定了extra_layer，extra_layer是一个 python list - 这两个条件一个为
True，一个为False与之后，会走到else分枝将 extra_layer 清空
lcy-seso
force-pushed
the
fix_config_parsing_bug
branch
from
d1e6e77
to
1e21259
Compare
lcy-seso
force-pushed
the
fix_config_parsing_bug
branch
from
ff71f9d
to
35332a2
Compare
emailweixu
reviewed
Jun 7, 2017
python/paddle/v2/layer.py
Outdated
| for l in cp.g_config.model_config.layers: | ||
| if l.name not in layer_names: | ||
| continue | ||
| model_config.layers.extend([l]) | ||
| if l.type == 'data': | ||
| if l.name in model_config.output_layer_names: | ||
| continue |
There was a problem hiding this comment.
- In text generation, the
outlinkto save the generated word indices is aDataLayer
https://github.com/PaddlePaddle/Paddle/blob/develop/python/paddle/trainer/config_parser.py#L427 defined in theRecurrentLayerGroupEndfunction. - If this
DataLayeris not excluded from inputs of the network, an error occurs during data feeding. - This
DataLayeris sure to be the output in text generation, so I add this statement.
There was a problem hiding this comment.
Need to add a comment. Otherwise it's very hard to understand.
python/paddle/v2/topology.py
Outdated
| return [(nm, data_layers[nm].data_type) | ||
| for nm in self.proto().input_layer_names] | ||
| for nm in self.proto().input_layer_names if nm in data_layers] |
There was a problem hiding this comment.
when is input_layer not a data_layer?
There was a problem hiding this comment.
This is a mistake. I have fixed this.
lcy-seso
commented
Jun 8, 2017
python/paddle/v2/layer.py
Outdated
| for l in cp.g_config.model_config.layers: | ||
| if l.name not in layer_names: | ||
| continue | ||
| model_config.layers.extend([l]) | ||
| if l.type == 'data': | ||
| if l.name in model_config.output_layer_names: | ||
| continue |
There was a problem hiding this comment.
- In text generation, the
outlinkto save the generated word indices is aDataLayer
https://github.com/PaddlePaddle/Paddle/blob/develop/python/paddle/trainer/config_parser.py#L427 defined in theRecurrentLayerGroupEndfunction. - If this
DataLayeris not excluded from inputs of the network, an error occurs during data feeding. - This
DataLayeris sure to be the output in text generation, so I add this statement.
python/paddle/v2/topology.py
Outdated
| return [(nm, data_layers[nm].data_type) | ||
| for nm in self.proto().input_layer_names] | ||
| for nm in self.proto().input_layer_names if nm in data_layers] |
There was a problem hiding this comment.
This is a mistake. I have fixed this.
lcy-seso
force-pushed
the
fix_config_parsing_bug
branch
from
48ad842
to
892b1f9
Compare
emailweixu
approved these changes
Jun 12, 2017
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
fix #2349. fix #2385
eos_layeris not trimmed as a used layer in text generation.dropout_layerfrompaddle.networktopaddle.layerStaticInputandGeneratedInput. Currently,StaticInputV2andGeneratedInputV2are not available anymore, useStaticInputandGeneratedInputinstead.