* Add sent_starts to GoldParse
* Add SentTagger pipeline component
Add `SentTagger` pipeline component as a subclass of `Tagger`.
* Model reduces default parameters from `Tagger` to be small and fast
* Hard-coded set of two labels:
* S (1): token at beginning of sentence
* I (0): all other sentence positions
* Sets `token.sent_start` values
* Add sentence segmentation to Scorer
Report `sent_p/r/f` for sentence boundaries, which may be provided by
various pipeline components.
* Add sentence segmentation to CLI evaluate
* Add senttagger metrics/scoring to train CLI
* Rename SentTagger to SentenceRecognizer
* Add SentenceRecognizer to spacy.pipes imports
* Add SentenceRecognizer serialization test
* Shorten component name to sentrec
* Remove duplicates from train CLI output metrics
* Switch to train_dataset() function in train CLI
* Fixes for pipe() methods in pipeline components
* Don't clobber `examples` variable with `as_example` in pipe() methods
* Remove unnecessary traversals of `examples`
* Update Parser.pipe() for Examples
* Add `as_examples` kwarg to `pipe()` with implementation to return
`Example`s
* Accept `Doc` or `Example` in `pipe()` with `_get_doc()` (copied from
`Pipe`)
* Fixes to Example implementation in spacy.gold
* Move `make_projective` from an attribute of Example to an argument of
`Example.get_gold_parses()`
* Head of 0 are not treated as unset
* Unset heads are set to self rather than `None` (which causes problems
while projectivizing)
* Check for `Doc` (not just not `None`) when creating GoldParses for
pre-merged example
* Don't clobber `examples` variable in `iter_gold_docs()`
* Add/modify gold tests for handling projectivity
* In JSON roundtrip compare results from `dev_dataset` rather than
`train_dataset` to avoid projectivization (and other potential
modifications)
* Add test for projective train vs. nonprojective dev versions of the
same `Doc`
* Handle ignore_misaligned as arg rather than attr
Move `ignore_misaligned` from an attribute of `Example` to an argument
to `Example.get_gold_parses()`, which makes it parallel to
`make_projective`.
Add test with old and new align that checks whether `ignore_misaligned`
errors are raised as expected (only for new align).
* Remove unused attrs from gold.pxd
Remove `ignore_misaligned` and `make_projective` from `gold.pxd`
* Refer to Example.goldparse in iter_gold_docs()
Use `Example.goldparse` in `iter_gold_docs()` instead of `Example.gold`
because a `None` `GoldParse` is generated with ignore_misaligned and
generating it on-the-fly can raise an unwanted AlignmentError
* Update test for ignore_misaligned
* Generalize handling of tokenizer special cases
Handle tokenizer special cases more generally by using the Matcher
internally to match special cases after the affix/token_match
tokenization is complete.
Instead of only matching special cases while processing balanced or
nearly balanced prefixes and suffixes, this recognizes special cases in
a wider range of contexts:
* Allows arbitrary numbers of prefixes/affixes around special cases
* Allows special cases separated by infixes
Existing tests/settings that couldn't be preserved as before:
* The emoticon '")' is no longer a supported special case
* The emoticon ':)' in "example:)" is a false positive again
When merged with #4258 (or the relevant cache bugfix), the affix and
token_match properties should be modified to flush and reload all
special cases to use the updated internal tokenization with the Matcher.
* Remove accidentally added test case
* Really remove accidentally added test
* Reload special cases when necessary
Reload special cases when affixes or token_match are modified. Skip
reloading during initialization.
* Update error code number
* Fix offset and whitespace in Matcher special cases
* Fix offset bugs when merging and splitting tokens
* Set final whitespace on final token in inserted special case
* Improve cache flushing in tokenizer
* Separate cache and specials memory (temporarily)
* Flush cache when adding special cases
* Repeated `self._cache = PreshMap()` and `self._specials = PreshMap()`
are necessary due to this bug:
https://github.com/explosion/preshed/issues/21
* Remove reinitialized PreshMaps on cache flush
* Update UD bin scripts
* Update imports for `bin/`
* Add all currently supported languages
* Update subtok merger for new Matcher validation
* Modify blinded check to look at tokens instead of lemmas (for corpora
with tokens but not lemmas like Telugu)
* Use special Matcher only for cases with affixes
* Reinsert specials cache checks during normal tokenization for special
cases as much as possible
* Additionally include specials cache checks while splitting on infixes
* Since the special Matcher needs consistent affix-only tokenization
for the special cases themselves, introduce the argument
`with_special_cases` in order to do tokenization with or without
specials cache checks
* After normal tokenization, postprocess with special cases Matcher for
special cases containing affixes
* Replace PhraseMatcher with Aho-Corasick
Replace PhraseMatcher with the Aho-Corasick algorithm over numpy arrays
of the hash values for the relevant attribute. The implementation is
based on FlashText.
The speed should be similar to the previous PhraseMatcher. It is now
possible to easily remove match IDs and matches don't go missing with
large keyword lists / vocabularies.
Fixes#4308.
* Restore support for pickling
* Fix internal keyword add/remove for numpy arrays
* Add test for #4248, clean up test
* Improve efficiency of special cases handling
* Use PhraseMatcher instead of Matcher
* Improve efficiency of merging/splitting special cases in document
* Process merge/splits in one pass without repeated token shifting
* Merge in place if no splits
* Update error message number
* Remove UD script modifications
Only used for timing/testing, should be a separate PR
* Remove final traces of UD script modifications
* Update UD bin scripts
* Update imports for `bin/`
* Add all currently supported languages
* Update subtok merger for new Matcher validation
* Modify blinded check to look at tokens instead of lemmas (for corpora
with tokens but not lemmas like Telugu)
* Add missing loop for match ID set in search loop
* Remove cruft in matching loop for partial matches
There was a bit of unnecessary code left over from FlashText in the
matching loop to handle partial token matches, which we don't have with
PhraseMatcher.
* Replace dict trie with MapStruct trie
* Fix how match ID hash is stored/added
* Update fix for match ID vocab
* Switch from map_get_unless_missing to map_get
* Switch from numpy array to Token.get_struct_attr
Access token attributes directly in Doc instead of making a copy of the
relevant values in a numpy array.
Add unsatisfactory warning for hash collision with reserved terminal
hash key. (Ideally it would change the reserved terminal hash and redo
the whole trie, but for now, I'm hoping there won't be collisions.)
* Restructure imports to export find_matches
* Implement full remove()
Remove unnecessary trie paths and free unused maps.
Parallel to Matcher, raise KeyError when attempting to remove a match ID
that has not been added.
* Switch to PhraseMatcher.find_matches
* Switch to local cdef functions for span filtering
* Switch special case reload threshold to variable
Refer to variable instead of hard-coded threshold
* Move more of special case retokenize to cdef nogil
Move as much of the special case retokenization to nogil as possible.
* Rewrap sort as stdsort for OS X
* Rewrap stdsort with specific types
* Switch to qsort
* Fix merge
* Improve cmp functions
* Fix realloc
* Fix realloc again
* Initialize span struct while retokenizing
* Temporarily skip retokenizing
* Revert "Move more of special case retokenize to cdef nogil"
This reverts commit 0b7e52c797.
* Revert "Switch to qsort"
This reverts commit a98d71a942.
* Fix specials check while caching
* Modify URL test with emoticons
The multiple suffix tests result in the emoticon `:>`, which is now
retokenized into one token as a special case after the suffixes are
split off.
* Refactor _apply_special_cases()
* Use cdef ints for span info used in multiple spots
* Modify _filter_special_spans() to prefer earlier
Parallel to #4414, modify _filter_special_spans() so that the earlier
span is preferred for overlapping spans of the same length.
* Replace MatchStruct with Entity
Replace MatchStruct with Entity since the existing Entity struct is
nearly identical.
* Replace Entity with more general SpanC
* Replace MatchStruct with SpanC
* Add error in debug-data if no dev docs are available (see #4575)
* Update azure-pipelines.yml
* Revert "Update azure-pipelines.yml"
This reverts commit ed1060cf59.
* Use latest wasabi
* Reorganise install_requires
* add dframcy to universe.json (#4580)
* Update universe.json [ci skip]
* Fix multiprocessing for as_tuples=True (#4582)
* Fix conllu script (#4579)
* force extensions to avoid clash between example scripts
* fix arg order and default file encoding
* add example config for conllu script
* newline
* move extension definitions to main function
* few more encodings fixes
* Add load_from_docbin example [ci skip]
TODO: upload the file somewhere
* Update README.md
* Add warnings about 3.8 (resolves#4593) [ci skip]
* Fixed typo: Added space between "recognize" and "various" (#4600)
* Fix DocBin.merge() example (#4599)
* Replace function registries with catalogue (#4584)
* Replace functions registries with catalogue
* Update __init__.py
* Fix test
* Revert unrelated flag [ci skip]
* Bugfix/dep matcher issue 4590 (#4601)
* add contributor agreement for prilopes
* add test for issue #4590
* fix on_match params for DependencyMacther (#4590)
* Minor updates to language example sentences (#4608)
* Add punctuation to Spanish example sentences
* Combine multilanguage examples for lang xx
* Add punctuation to nb examples
* Always realloc to a larger size
Avoid potential (unlikely) edge case and cymem error seen in #4604.
* Add error in debug-data if no dev docs are available (see #4575)
* Update debug-data for GoldCorpus / Example
* Ignore None label in misaligned NER data
* OrigAnnot class instead of gold.orig_annot list of zipped tuples
* from_orig to replace from_annot_tuples
* rename to RawAnnot
* some unit tests for GoldParse creation and internal format
* removing orig_annot and switching to lists instead of tuple
* rewriting tuples to use RawAnnot (+ debug statements, WIP)
* fix pop() changing the data
* small fixes
* pop-append fixes
* return RawAnnot for existing GoldParse to have uniform interface
* clean up imports
* fix merge_sents
* add unit test for 4402 with new structure (not working yet)
* introduce DocAnnot
* typo fixes
* add unit test for merge_sents
* rename from_orig to from_raw
* fixing unit tests
* fix nn parser
* read_annots to produce text, doc_annot pairs
* _make_golds fix
* rename golds_to_gold_annots
* small fixes
* fix encoding
* have golds_to_gold_annots use DocAnnot
* missed a spot
* merge_sents as function in DocAnnot
* allow specifying only part of the token-level annotations
* refactor with Example class + underlying dicts
* pipeline components to work with Example objects (wip)
* input checking
* fix yielding
* fix calls to update
* small fixes
* fix scorer unit test with new format
* fix kwargs order
* fixes for ud and conllu scripts
* fix reading data for conllu script
* add in proper errors (not fixed numbering yet to avoid merge conflicts)
* fixing few more small bugs
* fix EL script
* Flag to ignore examples with mismatched raw/gold text
After #4525, we're seeing some alignment failures on our OntoNotes data. I think we actually have fixes for most of these cases.
In general it's better to fix the data, but it seems good to allow the GoldCorpus class to just skip cases where the raw text doesn't
match up to the gold words. I think previously we were silently ignoring these cases.
* Try to fix test on Python 2.7
* 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
* Add early stopping
* Add return_score option to evaluate
* Fix missing str to path conversion
* Fix import + old python compatibility
* Fix bad beam_width setting during cpu evaluation in spacy train with gpu option turned on
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
* Add todo
* Auto-format
* Update wasabi pin
* Format training results with wasabi
* Remove loading animation from model saving
Currently behaves weirdly
* Inline messages
* Remove unnecessary path2str
Already taken care of by printer
* Inline messages in CLI
* Remove unused function
* Move loading indicator into loading function
* Check for invalid whitespace entities
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
