Custom space symbol #83
Open
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
specify the blank symbol with. The default blank symbol is " ". This was previously hard-coded in the decoder, causing problems when using it for handwriting recognition where the blank symbol may be different, for example "|". modified: ctcdecode/__init__.py modified: ctcdecode/src/binding.cpp modified: ctcdecode/src/binding.h modified: ctcdecode/src/ctc_beam_search_decoder.cpp modified: ctcdecode/src/ctc_beam_search_decoder.h modified: ctcdecode/src/decoder_utils.cpp modified: ctcdecode/src/scorer.cpp modified: ctcdecode/src/scorer.h
…be called on the space symbol before passing when passing it as an argument to "ctc_decode.paddle_get_scorer" and "ctc_decode.paddle_beam_decode" and "ctc_decode.paddle_beam_decode_lm". Otherwise it will not be in the const* char format expected by the c++ interface for these parameters. modified: ctcdecode/__init__.py
…ustom_blank_symbol
ryanleary
requested changes
Oct 29, 2018
| @@ -35,7 +36,9 @@ std::vector<std::pair<double, Output>> ctc_beam_search_decoder( | |||
| // size_t blank_id = vocabulary.size(); | |||
|
|
|||
| // assign space id | |||
| auto it = std::find(vocabulary.begin(), vocabulary.end(), " "); | |||
| // Changed by Gideon from the blank symbol " " to a custom symbol specified as argument | |||
There was a problem hiding this comment.
Dead code tells no lies. Please remove this comment and the line of code you commented out below (line 41). That's what git history is for.
| @@ -153,7 +153,8 @@ bool add_word_to_dictionary( | |||
| std::vector<int> int_word; | |||
|
|
|||
| for (auto &c : characters) { | |||
| if (c == " ") { | |||
| // if (c == " ") { | |||
| if (c == "|") { // Gideon: replaced the space symbol " " => "|" | |||
There was a problem hiding this comment.
This can't be hardcoded. You'll have to parameterize the space character. Though, looking at it more closely, it looks like you could probably just do a lookup based on the SPACE_ID param...
| @@ -16,7 +16,8 @@ using namespace lm::ngram; | |||
| Scorer::Scorer(double alpha, | |||
| double beta, | |||
| const std::string& lm_path, | |||
| const std::vector<std::string>& vocab_list) { | |||
| const std::vector<std::string>& vocab_list, | |||
| const std::string &space_symbol) { | |||
There was a problem hiding this comment.
Can you please set a default arg here.
| if (labels.empty()) return {}; | ||
|
|
||
| std::string s = vec2str(labels); | ||
| std::vector<std::string> words; | ||
| if (is_character_based_) { | ||
| words = split_utf8_str(s); | ||
| } else { | ||
| words = split_str(s, " "); | ||
| // words = split_str(s, " "); | ||
| words = split_str(s, space_symbol); //Gideon: replaced the space character from " " to a custom string |
There was a problem hiding this comment.
Delete dead code and comment
| if (char_list_[i] == " ") { | ||
| SPACE_ID_ = i; | ||
| //if (char_list_[i] == " ") { | ||
| if (char_list_[i] == space_symbol) { //Gideon: replaced the space character from " " to a custom string |
There was a problem hiding this comment.
Delete dead code and comment
…h, which facilitates usage with pytorch 1.0. Merge branch 'master' of https://github.com/parlance/ctcdecode into custom_space_symbol ctcdecode/__init__.py ctcdecode/src/binding.cpp ctcdecode/src/binding.h ctcdecode/src/ctc_beam_search_decoder.cpp ctcdecode/src/ctc_beam_search_decoder.h Changes to be committed: modified: README.md modified: build.py modified: ctcdecode/__init__.py modified: ctcdecode/src/binding.cpp modified: ctcdecode/src/binding.h modified: ctcdecode/src/ctc_beam_search_decoder.cpp modified: ctcdecode/src/ctc_beam_search_decoder.h modified: ctcdecode/src/decoder_utils.cpp modified: ctcdecode/src/decoder_utils.h modified: ctcdecode/src/path_trie.cpp modified: ctcdecode/src/path_trie.h modified: requirements.txt modified: setup.py modified: tests/test.py
"self._space_symbol.encode()" was swapped during merging of the code. - self._cutoff_prob, self.cutoff_top_n, self._blank_id,self._log_probs, self._space_symbol.encode(), output, timesteps, - scores, out_seq_len) + self._cutoff_prob, self.cutoff_top_n, self._blank_id, self._space_symbol.encode(), self._log_probs, + output, timesteps, scores, out_seq_len) modified: ctcdecode/__init__.py
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.
I added an extra argument to the decoders to allow specification of a custom space symbol. Currently the space symbol used by the decoder is hard-coded to be " ". This is probably fine in most cases, but it does not work for example for my problem domain of handwriting recognition, in which the word separator can be a special symbol such as "|" and the normal space symbol " " may be not used at all.