The helper method state.B(1) gets the index of the first token of the
buffer, or -1 if no such token exists. Normally this is safe because we
pass this to functions like state.safe_get(), which returns an empty
token. Here we used it directly as an array index, which is not okay!
This error may have been the cause of out-of-bounds access errors during
training. Similar errors may still be around, so much be hunted down.
Hunting this one down took a long time...I printed out values across
training runs and diffed, looking for points of divergence between
runs, when no randomness should be allowed.
The set_children_from_heads function assumed parse trees were
projective. However, non-projective parses may be passed in during
deserialization, or after deprojectivising. This caused incorrect
sentence boundaries to be set for non-projective parses. Close#2772.
* subword_features: Controls whether subword features are used in the
word embeddings. True by default (specifically, prefix, suffix and word
shape). Should be set to False for languages like Chinese and Japanese.
* conv_depth: Depth of the convolutional layers. Defaults to 4.
The parser.begin_training() method was rewritten in v2.1. The rewrite
introduced a regression, where if you added labels prior to
begin_training(), these labels were discarded. This patch fixes that.
Our JSON training format is annoying to work with, and we've wanted to
retire it for some time. In the meantime, we can at least add some
missing functions to make it easier to live with.
This patch adds a function that generates the JSON format from a list
of Doc objects, one per paragraph. This should be a convenient way to handle
a lot of data conversions: whatever format you have the source
information in, you can use it to setup a Doc object. This approach
should offer better future-proofing as well. Hopefully, we can steadily
rewrite code that is sensitive to the current data-format, so that it
instead goes through this function. Then when we change the data format,
we won't have such a problem.
I have added numbers in hindi lex_attrs.py file according to Indian numbering system(https://en.wikipedia.org/wiki/Indian_numbering_system) and here are there english translations:
'शून्य' => zero
'एक' => one
'दो' => two
'तीन' => three
'चार' => four
'पांच' => five
'छह' => six
'सात'=>seven
'आठ' => eight
'नौ' => nine
'दस' => ten
'ग्यारह' => eleven
'बारह' => twelve
'तेरह' => thirteen
'चौदह' => fourteen
'पंद्रह' => fifteen
'सोलह'=> sixteen
'सत्रह' => seventeen
'अठारह' => eighteen
'उन्नीस' => nineteen
'बीस' => twenty
'तीस' => thirty
'चालीस' => forty
'पचास' => fifty
'साठ' => sixty
'सत्तर' => seventy
'अस्सी' => eighty
'नब्बे' => ninety
'सौ' => hundred
'हज़ार' => thousand
'लाख' => hundred thousand
'करोड़' => ten million
'अरब' => billion
'खरब' => hundred billion
<!--- Provide a general summary of your changes in the title. -->
## Description
<!--- Use this section to describe your changes. If your changes required
testing, include information about the testing environment and the tests you
ran. If your test fixes a bug reported in an issue, don't forget to include the
issue number. If your PR is still a work in progress, that's totally fine – just
include a note to let us know. -->
### Types of change
<!-- What type of change does your PR cover? Is it a bug fix, an enhancement
or new feature, or a change to the documentation? -->
## Checklist
<!--- Before you submit the PR, go over this checklist and make sure you can
tick off all the boxes. [] -> [x] -->
- [x] I have submitted the spaCy Contributor Agreement.
- [x] I ran the tests, and all new and existing tests passed.
- [x] My changes don't require a change to the documentation, or if they do, I've added all required information.
* Exceptions for single letter words ending sentence
Sentences ending in "i." (as in "... peka i."), "m." (as in "...än 2000 m."), should be tokenized as two separate tokens.
* Add test
## Description
Related issues: #2379 (should be fixed by separating model tests)
* **total execution time down from > 300 seconds to under 60 seconds** 🎉
* removed all model-specific tests that could only really be run manually anyway – those will now live in a separate test suite in the [`spacy-models`](https://github.com/explosion/spacy-models) repository and are already integrated into our new model training infrastructure
* changed all relative imports to absolute imports to prepare for moving the test suite from `/spacy/tests` to `/tests` (it'll now always test against the installed version)
* merged old regression tests into collections, e.g. `test_issue1001-1500.py` (about 90% of the regression tests are very short anyways)
* tidied up and rewrote existing tests wherever possible
### Todo
- [ ] move tests to `/tests` and adjust CI commands accordingly
- [x] move model test suite from internal repo to `spacy-models`
- [x] ~~investigate why `pipeline/test_textcat.py` is flakey~~
- [x] review old regression tests (leftover files) and see if they can be merged, simplified or deleted
- [ ] update documentation on how to run tests
### Types of change
enhancement, tests
## Checklist
<!--- Before you submit the PR, go over this checklist and make sure you can
tick off all the boxes. [] -> [x] -->
- [x] I have submitted the spaCy Contributor Agreement.
- [x] I ran the tests, and all new and existing tests passed.
- [ ] My changes don't require a change to the documentation, or if they do, I've added all required information.
This patch improves tokenizer speed by about 10%, and reduces memory usage in the `Vocab` by removing a redundant index. The `vocab._by_orth` and `vocab._by_hash` indexed on different data in v1, but in v2 the orth and the hash are identical.
The patch also fixes an uninitialized variable in the tokenizer, the `has_special` flag. This checks whether a chunk we're tokenizing triggers a special-case rule. If it does, then we avoid caching within the chunk. This check led to incorrectly rejecting some chunks from the cache.
With the `en_core_web_md` model, we now tokenize the IMDB train data at 503,104k words per second. Prior to this patch, we had 465,764k words per second.
Before switching to the regex library and supporting more languages, we had 1.3m words per second for the tokenizer. In order to recover the missing speed, we need to:
* Fix the variable-length lookarounds in the suffix, infix and `token_match` rules
* Improve the performance of the `token_match` regex
* Switch back from the `regex` library to the `re` library.
## Checklist
<!--- Before you submit the PR, go over this checklist and make sure you can
tick off all the boxes. [] -> [x] -->
- [x] I have submitted the spaCy Contributor Agreement.
- [x] I ran the tests, and all new and existing tests passed.
- [x] My changes don't require a change to the documentation, or if they do, I've added all required information.
This jargon is not offencive but emotionally colored as funny due to its deviation from the norm for various reasons: immitating a dialect, deliberately wrong spelling emphasizing its low colloquial nature, obsolete form, foreign borrowing with native flections, etc.
Dmitry Briukhanov, Linguist & Pythonist
