Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add conv_seq_to_seq #431

Merged
merged 5 commits into from
Nov 14, 2017
Merged

Add conv_seq_to_seq #431

merged 5 commits into from
Nov 14, 2017

Conversation

ranqiu92
Copy link
Contributor

@ranqiu92 ranqiu92 commented Nov 7, 2017

Copy link
Collaborator

@lcy-seso lcy-seso left a comment

Choose a reason for hiding this comment

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

Some comments first, thank you for the work

:type drop_rate: float
:param max_len: The maximum length of the sentence to be generated.
:type max_len: int
:param beam_size: The width of beam search.
Copy link
Collaborator

Choose a reason for hiding this comment

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

The width of beam expansion.

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Done

:type input: LayerOutput
:param size: The dimension of the block's output.
:type size: int
:param context_len: The context width of the convolution.
Copy link
Collaborator

Choose a reason for hiding this comment

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

width --> length

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Done

:type token_emb: LayerOutput
:param pos_emb: The embedding vector of the input token's position.
:type pos_emb: LayerOutput
:param conv_blocks: The scale list of the convolution blocks. And each element of the
Copy link
Collaborator

Choose a reason for hiding this comment

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

And each element of the --> Each element of the

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Done

act=paddle.activation.Linear(),
param_attr=paddle.attr.Param(learning_rate=1.0 /
(2.0 * num_attention)),
bias_attr=True)
Copy link
Collaborator

Choose a reason for hiding this comment

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

Personally, I prefer to change res into residual, or shortcut. Am I right?

I think for quite a while and then realize this is a residual connection. I think res is not a quite common abbreviation. A meaningful name will make the topology easier to understand.

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Done

act=paddle.activation.Linear(),
bias_attr=True)

# TODO scaled by length
Copy link
Collaborator

Choose a reason for hiding this comment

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

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Oh, sorry, I just forgot to delete it. It can't be realized with Paddle in current version.

trg_next_word = trg_next_word + [trg_dict['<e>']] * padding_num
yield src_word, src_word_pos, trg_word, trg_word_pos, trg_next_word

return reader
Copy link
Collaborator

Choose a reason for hiding this comment

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

Could you please fix the non-printable character at the end of the line.

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Done

convolution block.
:type enc_conv_blocks: list of tuple
:param dec_conv_blocks: The scale list of the decoder's convolution blocks. And each element of
the list contains output dimension and context length of the corresponding
Copy link
Collaborator

Choose a reason for hiding this comment

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

This line is too long.

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Done

"--beam_size",
default=1,
type=int,
help="Beam search width. (default: %(default)s)")
Copy link
Collaborator

Choose a reason for hiding this comment

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

The width of beam expansion.

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Done

"--model_path",
type=str,
required=True,
help="Model path. (default: %(default)s)")
Copy link
Collaborator

Choose a reason for hiding this comment

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

The path of trained model.

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Done



if __name__ == '__main__':
main()
Copy link
Collaborator

Choose a reason for hiding this comment

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

Could you please fix the non-printable character at the end of the line.

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Done @lcy-seso


# Notes

Currently, the beam search will forward the whole network when predicting every word, which is a waste of time. And we will fix it later.
Copy link
Collaborator

Choose a reason for hiding this comment

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

beam search will forward the encoder multiple times when predicting each target word, which requires extra computations.

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Done @lcy-seso

Copy link
Collaborator

@lcy-seso lcy-seso left a comment

Choose a reason for hiding this comment

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

LGTM, thank you.

Copy link
Collaborator

@lcy-seso lcy-seso left a comment

Choose a reason for hiding this comment

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

LGTM, thank you for this work and please keep on optimizing this.

@lcy-seso lcy-seso merged commit 1edc831 into PaddlePaddle:develop Nov 14, 2017
@ranqiu92 ranqiu92 deleted the convseq2seq branch November 14, 2017 12:23
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.

None yet

2 participants