* Add doc.cats to spacy.gold at the paragraph level
Support `doc.cats` as `"cats": [{"label": string, "value": number}]` in
the spacy JSON training format at the paragraph level.
* `spacy.gold.docs_to_json()` writes `docs.cats`
* `GoldCorpus` reads in cats in each `GoldParse`
* Update instances of gold_tuples to handle cats
Update iteration over gold_tuples / gold_parses to handle addition of
cats at the paragraph level.
* Add textcat to train CLI
* Add textcat options to train CLI
* Add textcat labels in `TextCategorizer.begin_training()`
* Add textcat evaluation to `Scorer`:
* For binary exclusive classes with provided label: F1 for label
* For 2+ exclusive classes: F1 macro average
* For multilabel (not exclusive): ROC AUC macro average (currently
relying on sklearn)
* Provide user info on textcat evaluation settings, potential
incompatibilities
* Provide pipeline to Scorer in `Language.evaluate` for textcat config
* Customize train CLI output to include only metrics relevant to current
pipeline
* Add textcat evaluation to evaluate CLI
* Fix handling of unset arguments and config params
Fix handling of unset arguments and model confiug parameters in Scorer
initialization.
* Temporarily add sklearn requirement
* Remove sklearn version number
* Improve Scorer handling of models without textcats
* Fixing Scorer handling of models without textcats
* Update Scorer output for python 2.7
* Modify inf in Scorer for python 2.7
* Auto-format
Also make small adjustments to make auto-formatting with black easier and produce nicer results
* Move error message to Errors
* Update documentation
* Add cats to annotation JSON format [ci skip]
* Fix tpl flag and docs [ci skip]
* Switch to internal roc_auc_score
Switch to internal `roc_auc_score()` adapted from scikit-learn.
* Add AUCROCScore tests and improve errors/warnings
* Add tests for AUCROCScore and roc_auc_score
* Add missing error for only positive/negative values
* Remove unnecessary warnings and errors
* Make reduced roc_auc_score functions private
Because most of the checks and warnings have been stripped for the
internal functions and access is only intended through `ROCAUCScore`,
make the functions for roc_auc_score adapted from scikit-learn private.
* Check that data corresponds with multilabel flag
Check that the training instances correspond with the multilabel flag,
adding the multilabel flag if required.
* Add textcat score to early stopping check
* Add more checks to debug-data for textcat
* Add example training data for textcat
* Add more checks to textcat train CLI
* Check configuration when extending base model
* Fix typos
* Update textcat example data
* Provide licensing details and licenses for data
* Remove two labels with no positive instances from jigsaw-toxic-comment
data.
Co-authored-by: Ines Montani <ines@ines.io>
* Prevent subtok label if not learning tokens
The parser introduces the subtok label to mark tokens that should be
merged during post-processing. Previously this happened even if we did
not have the --learn-tokens flag set. This patch passes the config
through to the parser, to prevent the problem.
* Make merge_subtokens a parser post-process if learn_subtokens
* Fix train script
* Add test for 3830: subtok problem
* Fix handlign of non-subtok in parser training
* Improve error message when model.from_bytes() dies
When Thinc's model.from_bytes() is called with a mismatched model, often
we get a particularly ungraceful error,
e.g. "AttributeError: FunctionLayer has no attribute G"
This is because we're trying to load the parameters for something like
a LayerNorm layer, and the model architecture has some other layer there
instead. This is obviously terrible, especially since the error *type*
is wrong.
I've changed it to raise a ValueError. The error message is still
probably a bit terse, but it's hard to be sure exactly what's gone
wrong.
* Update spacy/pipeline/pipes.pyx
* Update spacy/pipeline/pipes.pyx
* Update spacy/pipeline/pipes.pyx
* Update spacy/syntax/nn_parser.pyx
* Update spacy/syntax/nn_parser.pyx
* Update spacy/pipeline/pipes.pyx
Co-Authored-By: Matthew Honnibal <honnibal+gh@gmail.com>
* Update spacy/pipeline/pipes.pyx
Co-Authored-By: Matthew Honnibal <honnibal+gh@gmail.com>
Co-authored-by: Ines Montani <ines@ines.io>
v2.1 introduced a regression when deserializing the parser after
parser.add_label() had been called. The code around the class mapping is
pretty confusing currently, as it was written to accommodate backwards
model compatibility. It needs to be revised when the models are next
retrained.
Closes#3433
* Make serialization methods consistent
exclude keyword argument instead of random named keyword arguments and deprecation handling
* Update docs and add section on serialization fields
* Improve handling of missing NER tags
GoldParse can accept missing NER tags, if entities is provided
in BILUO format (rather than as spans). Missing tags can be provided
as None values.
Fix bug that occurred when first tag was a None value. Closes#2603.
* Document specification of missing NER tags.
The output weights often return negative scores for classes, especially
via the bias terms. This means that when we add a new class, we can't
rely on just zeroing the weights, or we'll end up with positive
predictions for those labels.
To solve this, we use nan values as the initial weights for new labels.
This prevents them from ever coming out on top. During backprop, we
replace the nan values with the minimum assigned score, so that we're
still able to learn these classes.
After creating a component, the `.model` attribute is left with the value `True`, to indicate it should be created later during `from_disk()`, `from_bytes()` or `begin_training()`. This had led to confusing errors if you try to use the component without initializing the model.
To fix this, we add a method `require_model()` to the `Pipe` base class. The `require_model()` method needs to be called at the start of the `.predict()` and `.update()` methods of the components. It raises a `ValueError` if the model is not initialized. An error message has been added to `spacy.errors`.
The new spacy pretrain command implemented BERT/ULMFit/etc-like transfer learning, using our Language Modelling with Approximate Outputs version of BERT's cloze task. Pretraining is convenient, but in some ways it's a bit of a strange solution. All we're doing is initialising the weights. At the same time, we're putting a lot of work into our optimisation so that it's less sensitive to initial conditions, and more likely to find good optima. I discuss this a bit in the pseudo-rehearsal blog post: https://explosion.ai/blog/pseudo-rehearsal-catastrophic-forgetting
Support semi-supervised learning in spacy train
One obvious way to improve these pretraining methods is to do multi-task learning, instead of just transfer learning. This has been shown to work very well: https://arxiv.org/pdf/1809.08370.pdf . This patch makes it easy to do this sort of thing.
Add a new argument to spacy train, --raw-text. This takes a jsonl file with unlabelled data that can be used in arbitrary ways to do semi-supervised learning.
Add a new method to the Language class and to pipeline components, .rehearse(). This is like .update(), but doesn't expect GoldParse objects. It takes a batch of Doc objects, and performs an update on some semi-supervised objective.
Move the BERT-LMAO objective out from spacy/cli/pretrain.py into spacy/_ml.py, so we can create a new pipeline component, ClozeMultitask. This can be specified as a parser or NER multitask in the spacy train command. Example usage:
python -m spacy train en ./tmp ~/data/en-core-web/train/nw.json ~/data/en-core-web/dev/nw.json --pipeline parser --raw-textt ~/data/unlabelled/reddit-100k.jsonl --vectors en_vectors_web_lg --parser-multitasks cloze
Implement rehearsal methods for pipeline components
The new --raw-text argument and nlp.rehearse() method also gives us a good place to implement the the idea in the pseudo-rehearsal blog post in the parser. This works as follows:
Add a new nlp.resume_training() method. This allocates copies of pre-trained models in the pipeline, setting things up for the rehearsal updates. It also returns an optimizer object. This also greatly reduces confusion around the nlp.begin_training() method, which randomises the weights, making it not suitable for adding new labels or otherwise fine-tuning a pre-trained model.
Implement rehearsal updates on the Parser class, making it available for the dependency parser and NER. During rehearsal, the initial model is used to supervise the model being trained. The current model is asked to match the predictions of the initial model on some data. This minimises catastrophic forgetting, by keeping the model's predictions close to the original. See the blog post for details.
Implement rehearsal updates for tagger
Implement rehearsal updates for text categoriz
Remove hacks and wrappers, keep code in sync across our libraries and move spaCy a few steps closer to only depending on packages with binary wheels 🎉
See here: https://github.com/explosion/srsly
Serialization is hard, especially across Python versions and multiple platforms. After dealing with many subtle bugs over the years (encodings, locales, large files) our libraries like spaCy and Prodigy have steadily grown a number of utility functions to wrap the multiple serialization formats we need to support (especially json, msgpack and pickle). These wrapping functions ended up duplicated across our codebases, so we wanted to put them in one place.
At the same time, we noticed that having a lot of small dependencies was making maintainence harder, and making installation slower. To solve this, we've made srsly standalone, by including the component packages directly within it. This way we can provide all the serialization utilities we need in a single binary wheel.
srsly currently includes forks of the following packages:
ujson
msgpack
msgpack-numpy
cloudpickle
* WIP: replace json/ujson with srsly
* Replace ujson in examples
Use regular json instead of srsly to make code easier to read and follow
* Update requirements
* Fix imports
* Fix typos
* Replace msgpack with srsly
* Fix warning
* Create aryaprabhudesai.md (#2681)
* Update _install.jade (#2688)
Typo fix: "models" -> "model"
* Add FAC to spacy.explain (resolves#2706)
* Remove docstrings for deprecated arguments (see #2703)
* When calling getoption() in conftest.py, pass a default option (#2709)
* When calling getoption() in conftest.py, pass a default option
This is necessary to allow testing an installed spacy by running:
pytest --pyargs spacy
* Add contributor agreement
* update bengali token rules for hyphen and digits (#2731)
* Less norm computations in token similarity (#2730)
* Less norm computations in token similarity
* Contributor agreement
* Remove ')' for clarity (#2737)
Sorry, don't mean to be nitpicky, I just noticed this when going through the CLI and thought it was a quick fix. That said, if this was intention than please let me know.
* added contributor agreement for mbkupfer (#2738)
* Basic support for Telugu language (#2751)
* Lex _attrs for polish language (#2750)
* Signed spaCy contributor agreement
* Added polish version of english lex_attrs
* Introduces a bulk merge function, in order to solve issue #653 (#2696)
* Fix comment
* Introduce bulk merge to increase performance on many span merges
* Sign contributor agreement
* Implement pull request suggestions
* Describe converters more explicitly (see #2643)
* Add multi-threading note to Language.pipe (resolves#2582) [ci skip]
* Fix formatting
* Fix dependency scheme docs (closes#2705) [ci skip]
* Don't set stop word in example (closes#2657) [ci skip]
* Add words to portuguese language _num_words (#2759)
* Add words to portuguese language _num_words
* Add words to portuguese language _num_words
* Update Indonesian model (#2752)
* adding e-KTP in tokenizer exceptions list
* add exception token
* removing lines with containing space as it won't matter since we use .split() method in the end, added new tokens in exception
* add tokenizer exceptions list
* combining base_norms with norm_exceptions
* adding norm_exception
* fix double key in lemmatizer
* remove unused import on punctuation.py
* reformat stop_words to reduce number of lines, improve readibility
* updating tokenizer exception
* implement is_currency for lang/id
* adding orth_first_upper in tokenizer_exceptions
* update the norm_exception list
* remove bunch of abbreviations
* adding contributors file
* Fixed spaCy+Keras example (#2763)
* bug fixes in keras example
* created contributor agreement
* Adding French hyphenated first name (#2786)
* Fix typo (closes#2784)
* Fix typo (#2795) [ci skip]
Fixed typo on line 6 "regcognizer --> recognizer"
* Adding basic support for Sinhala language. (#2788)
* adding Sinhala language package, stop words, examples and lex_attrs.
* Adding contributor agreement
* Updating contributor agreement
* Also include lowercase norm exceptions
* Fix error (#2802)
* Fix error
ValueError: cannot resize an array that references or is referenced
by another array in this way. Use the resize function
* added spaCy Contributor Agreement
* Add charlax's contributor agreement (#2805)
* agreement of contributor, may I introduce a tiny pl languge contribution (#2799)
* Contributors agreement
* Contributors agreement
* Contributors agreement
* Add jupyter=True to displacy.render in documentation (#2806)
* Revert "Also include lowercase norm exceptions"
This reverts commit 70f4e8adf3.
* Remove deprecated encoding argument to msgpack
* Set up dependency tree pattern matching skeleton (#2732)
* Fix bug when too many entity types. Fixes#2800
* Fix Python 2 test failure
* Require older msgpack-numpy
* Restore encoding arg on msgpack-numpy
* Try to fix version pin for msgpack-numpy
* Update Portuguese Language (#2790)
* Add words to portuguese language _num_words
* Add words to portuguese language _num_words
* Portuguese - Add/remove stopwords, fix tokenizer, add currency symbols
* Extended punctuation and norm_exceptions in the Portuguese language
* Correct error in spacy universe docs concerning spacy-lookup (#2814)
* Update Keras Example for (Parikh et al, 2016) implementation (#2803)
* bug fixes in keras example
* created contributor agreement
* baseline for Parikh model
* initial version of parikh 2016 implemented
* tested asymmetric models
* fixed grevious error in normalization
* use standard SNLI test file
* begin to rework parikh example
* initial version of running example
* start to document the new version
* start to document the new version
* Update Decompositional Attention.ipynb
* fixed calls to similarity
* updated the README
* import sys package duh
* simplified indexing on mapping word to IDs
* stupid python indent error
* added code from https://github.com/tensorflow/tensorflow/issues/3388 for tf bug workaround
* Fix typo (closes#2815) [ci skip]
* Update regex version dependency
* Set version to 2.0.13.dev3
* Skip seemingly problematic test
* Remove problematic test
* Try previous version of regex
* Revert "Remove problematic test"
This reverts commit bdebbef455.
* Unskip test
* Try older version of regex
* 💫 Update training examples and use minibatching (#2830)
<!--- Provide a general summary of your changes in the title. -->
## Description
Update the training examples in `/examples/training` to show usage of spaCy's `minibatch` and `compounding` helpers ([see here](https://spacy.io/usage/training#tips-batch-size) for details). The lack of batching in the examples has caused some confusion in the past, especially for beginners who would copy-paste the examples, update them with large training sets and experienced slow and unsatisfying results.
### Types of change
enhancements
## 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.
* Visual C++ link updated (#2842) (closes#2841) [ci skip]
* New landing page
* Add contribution agreement
* Correcting lang/ru/examples.py (#2845)
* Correct some grammatical inaccuracies in lang\ru\examples.py; filled Contributor Agreement
* Correct some grammatical inaccuracies in lang\ru\examples.py
* Move contributor agreement to separate file
* Set version to 2.0.13.dev4
* Add Persian(Farsi) language support (#2797)
* Also include lowercase norm exceptions
* Remove in favour of https://github.com/explosion/spaCy/graphs/contributors
* Rule-based French Lemmatizer (#2818)
<!--- 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. -->
Add a rule-based French Lemmatizer following the english one and the excellent PR for [greek language optimizations](https://github.com/explosion/spaCy/pull/2558) to adapt the Lemmatizer class.
### 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? -->
- Lemma dictionary used can be found [here](http://infolingu.univ-mlv.fr/DonneesLinguistiques/Dictionnaires/telechargement.html), I used the XML version.
- Add several files containing exhaustive list of words for each part of speech
- Add some lemma rules
- Add POS that are not checked in the standard Lemmatizer, i.e PRON, DET, ADV and AUX
- Modify the Lemmatizer class to check in lookup table as a last resort if POS not mentionned
- Modify the lemmatize function to check in lookup table as a last resort
- Init files are updated so the model can support all the functionalities mentioned above
- Add words to tokenizer_exceptions_list.py in respect to regex used in tokenizer_exceptions.py
## 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.
* Set version to 2.0.13
* Fix formatting and consistency
* Update docs for new version [ci skip]
* Increment version [ci skip]
* Add info on wheels [ci skip]
* Adding "This is a sentence" example to Sinhala (#2846)
* Add wheels badge
* Update badge [ci skip]
* Update README.rst [ci skip]
* Update murmurhash pin
* Increment version to 2.0.14.dev0
* Update GPU docs for v2.0.14
* Add wheel to setup_requires
* Import prefer_gpu and require_gpu functions from Thinc
* Add tests for prefer_gpu() and require_gpu()
* Update requirements and setup.py
* Workaround bug in thinc require_gpu
* Set version to v2.0.14
* Update push-tag script
* Unhack prefer_gpu
* Require thinc 6.10.6
* Update prefer_gpu and require_gpu docs [ci skip]
* Fix specifiers for GPU
* Set version to 2.0.14.dev1
* Set version to 2.0.14
* Update Thinc version pin
* Increment version
* Fix msgpack-numpy version pin
* Increment version
* Update version to 2.0.16
* Update version [ci skip]
* Redundant ')' in the Stop words' example (#2856)
<!--- 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] -->
- [ ] I have submitted the spaCy Contributor Agreement.
- [ ] 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.
* Documentation improvement regarding joblib and SO (#2867)
Some documentation improvements
## Description
1. Fixed the dead URL to joblib
2. Fixed Stack Overflow brand name (with space)
### Types of change
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.
* raise error when setting overlapping entities as doc.ents (#2880)
* Fix out-of-bounds access in NER training
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.
* Change PyThaiNLP Url (#2876)
* Fix missing comma
* Add example showing a fix-up rule for space entities
* Set version to 2.0.17.dev0
* Update regex version
* Revert "Update regex version"
This reverts commit 62358dd867.
* Try setting older regex version, to align with conda
* Set version to 2.0.17
* Add spacy-js to universe [ci-skip]
* Add spacy-raspberry to universe (closes#2889)
* Add script to validate universe json [ci skip]
* Removed space in docs + added contributor indo (#2909)
* - removed unneeded space in documentation
* - added contributor info
* Allow input text of length up to max_length, inclusive (#2922)
* Include universe spec for spacy-wordnet component (#2919)
* feat: include universe spec for spacy-wordnet component
* chore: include spaCy contributor agreement
* Minor formatting changes [ci skip]
* Fix image [ci skip]
Twitter URL doesn't work on live site
* Check if the word is in one of the regular lists specific to each POS (#2886)
* 💫 Create random IDs for SVGs to prevent ID clashes (#2927)
Resolves#2924.
## Description
Fixes problem where multiple visualizations in Jupyter notebooks would have clashing arc IDs, resulting in weirdly positioned arc labels. Generating a random ID prefix so even identical parses won't receive the same IDs for consistency (even if effect of ID clash isn't noticable here.)
### Types of change
bug fix
## 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.
* Fix typo [ci skip]
* fixes symbolic link on py3 and windows (#2949)
* fixes symbolic link on py3 and windows
during setup of spacy using command
python -m spacy link en_core_web_sm en
closes#2948
* Update spacy/compat.py
Co-Authored-By: cicorias <cicorias@users.noreply.github.com>
* Fix formatting
* Update universe [ci skip]
* Catalan Language Support (#2940)
* Catalan language Support
* Ddding Catalan to documentation
* Sort languages alphabetically [ci skip]
* Update tests for pytest 4.x (#2965)
<!--- Provide a general summary of your changes in the title. -->
## Description
- [x] Replace marks in params for pytest 4.0 compat ([see here](https://docs.pytest.org/en/latest/deprecations.html#marks-in-pytest-mark-parametrize))
- [x] Un-xfail passing tests (some fixes in a recent update resolved a bunch of issues, but tests were apparently never updated here)
### 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.
* Fix regex pin to harmonize with conda (#2964)
* Update README.rst
* Fix bug where Vocab.prune_vector did not use 'batch_size' (#2977)
Fixes#2976
* Fix typo
* Fix typo
* Remove duplicate file
* Require thinc 7.0.0.dev2
Fixes bug in gpu_ops that would use cupy instead of numpy on CPU
* Add missing import
* Fix error IDs
* Fix tests
These experiments were completed a few weeks ago, but I didn't make the PR, pending model release.
Token vector width: 128->96
Hidden width: 128->64
Embed size: 5000->2000
Dropout: 0.2->0.1
Updated optimizer defaults (unclear how important?)
This should improve speed, model size and load time, while keeping
similar or slightly better accuracy.
The tl;dr is we prefer to prevent over-fitting by reducing model size,
rather than using more dropout.
Our epic matrix multiplication odyssey is drawing to a close...
I've now finally got the Blis linear algebra routines in a self-contained Python package, with wheels for Windows, Linux and OSX. The only missing platform at the moment is Windows Python 2.7. The result is at https://github.com/explosion/cython-blis
Thinc v7.0.0 will make the change to Blis. I've put a Thinc v7.0.0.dev0 up on PyPi so that we can test these changes with the CI, and even get them out to spacy-nightly, before Thinc v7.0.0 is released. This PR also updates the other dependencies to be in line with the current versions master is using. I've also resolved the msgpack deprecation problems, and gotten spaCy and Thinc up to date with the latest Cython.
The point of switching to Blis is to have control of how our matrix multiplications are executed across platforms. When we were using numpy for this, a different library would be used on pip and conda, OSX would use Accelerate, etc. This would open up different bugs and performance problems, especially when multi-threading was introduced.
With the change to Blis, we now strictly single-thread the matrix multiplications. This will make it much easier to use multiprocessing to parallelise the runtime, since we won't have nested parallelism problems to deal with.
* Use blis
* Use -2 arg to Cython
* Update dependencies
* Fix requirements
* Update setup dependencies
* Fix requirement typo
* Fix msgpack errors
* Remove Python27 test from Appveyor, until Blis works there
* Auto-format setup.py
* Fix murmurhash version
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 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.
adrianeboyd