* Test for arc levels for identical arcs
Also moves the test in order with the other numbered tests.
* displaCy: filter identical arcs
Avoid increased levels due to identical arcs by first
filtering any identical arcs.
* Sort keys before filtering
Manual entry with keys out of order would previously become
different tuples and therefore not filtered correctly.
Co-authored-by: Joachim Fainberg <joachimfainberg@Joachims-MBP.lan>
* Add basic tests for Tamil (ta)
* Add comment
Remove superfluous condition
* Remove superfluous call to `pipe`
Instantiate new tokenizer for special case
* Add initial design for diff command
For now, the diffing process looks like this:
- The default config is created based from some values in the user
config (e.g. which pipeline components were used, the lang, etc.)
- The user must supply manually if it was optimized for acc/efficiency
and if pretraining was involved.
* Make diff command structure similar to siblings
* Include gpu as a user option for CLI
* Make variables more explicit
* Fix type declaration for optimize enum
* Improve docstrings for diff CLI
* Add debug-diff to website API docs
* Switch position of configs so that user config is modded
* Add markdown flag for debug diff
This commit adds a --markdown (--md) flag that allows easier
copy-pasting to Github issues. Please note that this commit is dependent
on an unreleased version of wasabi (for the time being).
For posterity, the related PR is found here: https://github.com/ines/wasabi/pull/20
* Bump version of wasabi to 0.9.1
So that we can use the add_symbols parameter.
* Apply suggestions from code review
Co-authored-by: Ines Montani <ines@ines.io>
* Update docs based on code review suggestions
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
* Change command name from diff -> diff-config
* Clarify when options are relevant or not
* Rerun prettier on cli.md
Co-authored-by: Ines Montani <ines@ines.io>
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
* Added test for overlapping arcs
* Provide distinct levels to overlapping arcs
* Update return type hint for get_levels
* Improved formatting spacy/displacy/render.py
Co-authored-by: Ines Montani <ines@ines.io>
Co-authored-by: Joachim Fainberg <joachimfainberg@Joachims-MacBook-Pro.local>
Co-authored-by: Ines Montani <ines@ines.io>
* Added new convenience cython functions to SpanGroup to avoid unnecessary allocation/deallocation of objects
* Replaced sorting in has_overlap with C++ for efficiency. Also, added a test for has_overlap
* Added a method to efficiently merge SpanGroups
* Added __delitem__, __add__ and __iadd__. Also, allowed to pass span lists to merge function. Replaced extend() body with call to merge
* Renamed merge to concat and added missing things to documentation
* Added operator+ and operator += in the documentation
* Added a test for Doc deallocation
* Update spacy/tokens/span_group.pyx
* Updated SpanGroup tests to use new span list comparison function rather than assert_span_list_equal, eliminating the need to have a separate assert_not_equal fnction
* Fixed typos in SpanGroup documentation
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
* Minor changes requested by Sofie: rearranged import statements. Added new=3.2.1 tag to SpanGroup.__setitem__ documentation
* SpanGroup: moved repetitive list index check/adjustment in a separate function
* Turn off formatting that hurts readability spacy/tests/doc/test_span_group.py
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
* Remove formatting that hurts readability spacy/tests/doc/test_span_group.py
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
* Turn off formatting that hurts readability in spacy/tests/doc/test_span_group.py
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
* Support more internal methods for SpanGroup
Add support for:
* `__setitem__`
* `__delitem__`
* `__iadd__`: for `SpanGroup` or `Iterable[Span]`
* `__add__`: for `SpanGroup` only
Adapted from #9698 with the scope limited to the magic methods.
* Use v3.3 as new version in docs
* Add new tag to SpanGroup.copy in API docs
* Remove duplicate import
* Apply suggestions from code review
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
* Remaining suggestions and formatting
Co-authored-by: nrodnova <nrodnova@hotmail.com>
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
Co-authored-by: Natalia Rodnova <4512370+nrodnova@users.noreply.github.com>
* Alignment: use a simplified ragged type for performance
This introduces the AlignmentArray type, which is a simplified version
of Ragged that performs better on the simple(r) indexing performed for
alignment.
* AlignmentArray: raise an error when using unsupported index
* AlignmentArray: move error messages to Errors
* AlignmentArray: remove simlified ... with simplifications
* AlignmentArray: fix typo that broke a[n:n] indexing
* added failing test case for the issue.
* Fixed typo.
* fixed typo in test.
* added corrected typo word into test_tr_lex_attrs_capitals as param. Test passes. Also tried and confirmed that test is failing after fixing the typo in the test case I wrote. Deleted the test case for typo.
Co-authored-by: Yunus Atahan <yunus.atahan@trmotor.local>
* Add vector deduplication
* Add `Vocab.deduplicate_vectors()`
* Always run deduplication in `spacy init vectors`
* Clean up a few vector-related error messages and docs examples
* Always unique with numpy
* Fix types
* Add edit tree lemmatizer
Co-authored-by: Daniël de Kok <me@danieldk.eu>
* Hide edit tree lemmatizer labels
* Use relative imports
* Switch to single quotes in error message
* Type annotation fixes
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
* Reformat edit_tree_lemmatizer with black
* EditTreeLemmatizer.predict: take Iterable
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
* Validate edit trees during deserialization
This change also changes the serialized representation. Rather than
mirroring the deep C structure, we use a simple flat union of the match
and substitution node types.
* Move edit_trees to _edit_tree_internals
* Fix invalid edit tree format error message
* edit_tree_lemmatizer: remove outdated TODO comment
* Rename factory name to trainable_lemmatizer
* Ignore type instead of casting truths to List[Union[Ints1d, Floats2d, List[int], List[str]]] for thinc v8.0.14
* Switch to Tagger.v2
* Add documentation for EditTreeLemmatizer
* docs: Fix 3.2 -> 3.3 somewhere
* trainable_lemmatizer documentation fixes
* docs: EditTreeLemmatizer is in edit_tree_lemmatizer.py
Co-authored-by: Daniël de Kok <me@danieldk.eu>
Co-authored-by: Daniël de Kok <me@github.danieldk.eu>
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
* Added examples for Slovene
* Update spacy/lang/sl/examples.py
Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com>
* Corrected a typo in one of the sentences
Co-authored-by: Luka Dragar <D20124481@mytudublin.ie>
Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com>
* Add tokenizer option to allow Matcher handling for all rules
Add tokenizer option `with_faster_rules_heuristics` that determines
whether the special cases applied by the internal `Matcher` are filtered
by whether they contain affixes or space. If `True` (default), the rules
are filtered to prioritize speed over rare edge cases. If `False`, all
rules are included in the final `Matcher`-based pass over the doc.
* Reset all caches when reloading special cases
* Revert "Reset all caches when reloading special cases"
This reverts commit 4ef6bd171d.
* Initialize max_length properly
* Add new tag to API docs
* Rename to faster heuristics
* Save vectors as little endian, load with Ops.asarray
* Always save vector data as little endian
* Always run `Vectors.to_ops` when vector data is loaded so that
`Ops.asarray` can be used to load the data correctly for the current
ops.
* Update spacy/vectors.pyx
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
* Update spacy/vectors.pyx
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
* Fix docstring for EntityRenderer
* Add warning in displacy if doc.spans are empty
* Implement parse_spans converter
One notable change here is that the default spans_key is sc, and
it's set by the user through the options.
* Implement SpanRenderer
Here, I implemented a SpanRenderer that looks similar to the
EntityRenderer except for some templates. The spans_key, by default, is
set to sc, but can be configured in the options (see parse_spans). The
way I rendered these spans is per-token, i.e., I first check if each
token (1) belongs to a given span type and (2) a starting token of a
given span type. Once I have this information, I render them into the
markup.
* Fix mypy issues on typing
* Add tests for displacy spans support
* Update colors from RGB to hex
Co-authored-by: Ines Montani <ines@ines.io>
* Remove unnecessary CSS properties
* Add documentation for website
* Remove unnecesasry scripts
* Update wording on the documentation
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
* Put typing dependency on top of file
* Put back z-index so that spans overlap properly
* Make warning more explicit for spans_key
Co-authored-by: Ines Montani <ines@ines.io>
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
* Tagger: use unnormalized probabilities for inference
Using unnormalized softmax avoids use of the relatively expensive exp function,
which can significantly speed up non-transformer models (e.g. I got a speedup
of 27% on a German tagging + parsing pipeline).
* Add spacy.Tagger.v2 with configurable normalization
Normalization of probabilities is disabled by default to improve
performance.
* Update documentation, models, and tests to spacy.Tagger.v2
* Move Tagger.v1 to spacy-legacy
* docs/architectures: run prettier
* Unnormalized softmax is now a Softmax_v2 option
* Require thinc 8.0.14 and spacy-legacy 3.0.9
* Add save_candidates attribute
* Change spancat api
* Add unit test
* reimplement method to produce a list of doc
* Add method to docs
* Add new version tag
* Add intended use to docstring
* prettier formatting
* Add support basic support for lower sorbian.
* Add some test for dsb.
* Update spacy/lang/dsb/examples.py
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
* Fix get_matching_ents
Not sure what happened here - the code prior to this commit simply does
not work. It's already covered by entity linker tests, which were
succeeding in the NEL PR, but couldn't possibly succeed on master.
* Fix test
Test was indented inside another test and so doesn't seem to have been
running properly.
* Add support basic support for upper sorbian.
* Add tokenizer exceptions and tests.
* Update spacy/lang/hsb/examples.py
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
* Partial fix of entity linker batching
* Add import
* Better name
* Add `use_gold_ents` option, docs
* Change to v2, create stub v1, update docs etc.
* Fix error type
Honestly no idea what the right type to use here is.
ConfigValidationError seems wrong. Maybe a NotImplementedError?
* Make mypy happy
* Add hacky fix for init issue
* Add legacy pipeline entity linker
* Fix references to class name
* Add __init__.py for legacy
* Attempted fix for loss issue
* Remove placeholder V1
* formatting
* slightly more interesting train data
* Handle batches with no usable examples
This adds a test for batches that have docs but not entities, and a
check in the component that detects such cases and skips the update step
as thought the batch were empty.
* Remove todo about data verification
Check for empty data was moved further up so this should be OK now - the
case in question shouldn't be possible.
* Fix gradient calculation
The model doesn't know which entities are not in the kb, so it generates
embeddings for the context of all of them.
However, the loss does know which entities aren't in the kb, and it
ignores them, as there's no sensible gradient.
This has the issue that the gradient will not be calculated for some of
the input embeddings, which causes a dimension mismatch in backprop.
That should have caused a clear error, but with numpyops it was causing
nans to happen, which is another problem that should be addressed
separately.
This commit changes the loss to give a zero gradient for entities not in
the kb.
* add failing test for v1 EL legacy architecture
* Add nasty but simple working check for legacy arch
* Clarify why init hack works the way it does
* Clarify use_gold_ents use case
* Fix use gold ents related handling
* Add tests for no gold ents and fix other tests
* Use aligned ents function (not working)
This doesn't actually work because the "aligned" ents are gold-only. But
if I have a different function that returns the intersection, *then*
this will work as desired.
* Use proper matching ent check
This changes the process when gold ents are not used so that the
intersection of ents in the pred and gold is used.
* Move get_matching_ents to Example
* Use model attribute to check for legacy arch
* Rename flag
* bump spacy-legacy to lower 3.0.9
Co-authored-by: svlandeg <svlandeg@github.com>
* fixing argument order for rehearse
* rehearse test for ner and tagger
* rehearse bugfix
* added test for parser
* test for multilabel textcat
* rehearse fix
* remove debug line
* Update spacy/tests/training/test_rehearse.py
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
* Update spacy/tests/training/test_rehearse.py
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
Co-authored-by: Kádár Ákos <akos@onyx.uvt.nl>
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
* Make core projectivization methods cdef nogil
While profiling the parser, I noticed that relatively a lot of time is
spent in projectivization. This change rewrites the functions in the
core loops as cdef nogil for efficiency.
In C++-land, we use vector in place of Python lists and absent heads
are represented as -1 in place of None.
* _heads_to_c: add assertion
Validation should be performed by the caller, but this assertion ensures that
we are not reading/writing out of bounds with incorrect input.
* Fix NER check in CoNLL-U converter
Leave ents unset if no NER annotation is found in the MISC column.
* Revert to global rather than per-sentence NER check
* Update spacy/training/converters/conllu_to_docs.py
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
Add whitespace augmenter that inserts a single whitespace token into a
doc containing annotation used in core trained pipelines.
Add a combined augmenter that handles lowercasing, orth variants and
whitespace augmentation.
* Extended list of numbers for ru language
Extended list of numbers with all forms and cases including short forms, slang variants and roman numerals.
* Update lex_attrs.py
* Update 'like_num' function with percentages
Added support for numbers with percentages like 12%, 1.2% and etc. to the 'like_num' function.
* black formatting
Co-authored-by: thomashacker <EdwardSchmuhl@web.de>
* Extend list of abbreviations for ru language
Extended list of abbreviations for ru language those may have influence on tokenization.
* black formatting
Co-authored-by: thomashacker <EdwardSchmuhl@web.de>
* Delay loading of mecab in Korean tokenizer
Delay loading of mecab until the tokenizer is called the first time so
that it's possible to initialize a blank `ko` pipeline without having
mecab installed, e.g. for use with `spacy init vectors`.
* Move mecab import back to __init__
Move mecab import back to __init__ to warn users at the same point as
before for missing python dependencies.
* remove duplicate line
* add sent start/end token attributes to the docs
* let has_annotation work with IS_SENT_END
* elif instead of if
* add has_annotation test for sent attributes
* fix typo
* remove duplicate is_sent_start entry in docs
* Setup debug data for spancat
* Add check for missing labels
* Add low-level data warning error
* Improve logic when compiling the gold train data
* Implement check for negative examples
* Remove breakpoint
* Remove ws_ents and missing entity checks
* Fix mypy errors
* Make variable name spans_key consistent
* Rename pipeline -> component for consistency
* Account for missing labels per spans_key
* Cleanup variable names for consistency
* Improve brevity of conditional statements
* Remove unused variables
* Include spans_key as an argument for _get_examples
* Add a conditional check for spans_key
* Update spancat debug data based on new API
- Instead of using _get_labels_from_model(), I'm now using
_get_labels_from_spancat() (cf. https://github.com/explosion/spaCy/pull10079)
- The way information is displayed was also changed (text -> table)
* Rename model_labels to ensure mypy works
* Update wording on warning messages
Use "span type" instead of "entity type" in wording the warning messages.
This is because Spans aren't necessarily entities.
* Update component type into a Literal
This is to make it clear that the component parameter should only accept
either 'spancat' or 'ner'.
* Update checks to include actual model span_keys
Instead of looking at everything in the data, we only check those
span_keys from the actual spancat component. Instead of doing the filter
inside the for-loop, I just made another dictionary,
data_labels_in_component to hold this value.
* Update spacy/cli/debug_data.py
* Show label counts only when verbose is True
Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com>
Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com>
* Fix debug data check for ents that cross sents
* Use aligned sent starts to have the same indices for the NER and sent
start annotation
* Add a temporary, insufficient hack for the case where a
sentence-initial reference token is split into multiple tokens in the
predicted doc, since `Example.get_aligned("SENT_START")` currently
aligns `True` to all the split tokens.
* Improve test example
* Use Example.get_aligned_sent_starts
* Add test for crossing entity
* Auto-format code with black
* add black requirement to dev dependencies and pin to 22.x
* ignore black dependency for comparison with setup.cfg
Co-authored-by: explosion-bot <explosion-bot@users.noreply.github.com>
Co-authored-by: svlandeg <svlandeg@github.com>
So that overriding `paths.vectors` works consistently in generated
configs, set vectors model in `paths.vectors` and always refer to this
path in `initialize.vectors`.
Remove exception for whitespace tokens in `Example.get_aligned` so that
annotation on whitespace tokens is aligned in the same way as for
non-whitespace tokens.
* Clarify Span.ents documentation
Ref: #10135
Retain current behaviour. Span.ents will only include entities within
said span. You can't get tokens outside of the original span.
* Reword docstrings
Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com>
* Update API docs in the website
Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com>
* This comma has been most probably been left out unintentionally, leading to string concatenation between the two consecutive lines. This issue has been found automatically using a regular expression.
* This comma has been most probably been left out unintentionally, leading to string concatenation between the two consecutive lines. This issue has been found automatically using a regular expression.
* Fix infix as prefix in Tokenizer.explain
Update `Tokenizer.explain` to align with the `Tokenizer` algorithm:
* skip infix matches that are prefixes in the current substring
* Update tokenizer pseudocode in docs
* Improve typing hints for Matcher.__call__
* Add typing hints for DependencyMatcher
* Add typing hints to underscore extensions
* Update Doc.tensor type (requires numpy 1.21)
* Fix typing hints for Language.component decorator
* Use generic np.ndarray type in Doc to avoid numpy version update
* Fix mypy errors
* Fix cyclic import caused by Underscore typing hints
* Use Literal type from spacy.compat
* Update matcher.pyi import format
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
Instead of the running the actual suggester, which may require
annotation from annotating components that is not necessarily present in
the reference docs, use the built-in 1-gram suggester.
* added iob to int
* added tests
* added iob strings
* added error
* blacked attrs
* Update spacy/tests/lang/test_attrs.py
Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com>
* Update spacy/attrs.pyx
Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com>
* added iob strings as global
* minor refinement with iob
* removed iob strings from token
* changed to uppercase
* cleaned and went back to master version
* imported iob from attrs
* Update and format errors
* Support and test both str and int ENT_IOB key
Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com>
* added new field
* added exception for IOb strings
* minor refinement to schema
* removed field
* fixed typo
* imported numeriacla val
* changed the code bit
* cosmetics
* added test for matcher
* set ents of moc docs
* added invalid pattern
* minor update to documentation
* blacked matcher
* added pattern validation
* add IOB vals to schema
* changed into test
* mypy compat
* cleaned left over
* added compat import
* changed type
* added compat import
* changed literal a bit
* went back to old
* made explicit type
* Update spacy/schemas.py
Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com>
* Update spacy/schemas.py
Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com>
* Update spacy/schemas.py
Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com>
Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com>
* Determine labels by factory name in debug data
For all components, return labels for all components with the
corresponding factory name rather than for only the default name.
For `spancat`, return labels as a dict keyed by `spans_key`.
* Refactor for typing
* Add test
* Use assert instead of cast, removed unneeded arg
* Mark test as slow
* Use Vectors.shape rather than Vectors.data.shape
* Use Vectors.size rather than Vectors.data.size
* Add Vectors.to_ops to move data between different ops
* Add documentation for Vector.to_ops
* Corrected Span's __richcmp__ implementation to take end, label and kb_id in consideration
* Updated test
* Updated test
* Removed formatting from a test for readability sake
* Use same tuples for all comparisons
Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com>
* Edited Slovenian stop words list (#9707)
* Noun chunks for Italian (#9662)
* added it vocab
* copied portuguese
* added possessive determiner
* added conjed Nps
* added nmoded Nps
* test misc
* more examples
* fixed typo
* fixed parenth
* fixed comma
* comma fix
* added syntax iters
* fix some index problems
* fixed index
* corrected heads for test case
* fixed tets case
* fixed determiner gender
* cleaned left over
* added example with apostophe
* French NP review (#9667)
* adapted from pt
* added basic tests
* added fr vocab
* fixed noun chunks
* more examples
* typo fix
* changed naming
* changed the naming
* typo fix
* Add Japanese kana characters to default exceptions (fix#9693) (#9742)
This includes the main kana, or phonetic characters, used in Japanese.
There are some supplemental kana blocks in Unicode outside the BMP that
could also be included, but because their actual use is rare I omitted
them for now, but maybe they should be added. The omitted blocks are:
- Kana Supplement
- Kana Extended (A and B)
- Small Kana Extension
* Remove NER words from stop words in Norwegian (#9820)
Default stop words in Norwegian bokmål (nb) in Spacy contain important entities, e.g. France, Germany, Russia, Sweden and USA, police district, important units of time, e.g. months and days of the week, and organisations.
Nobody expects their presence among the default stop words. There is a danger of users complying with the general recommendation of filtering out stop words, while being unaware of filtering out important entities from their data.
See explanation in https://github.com/explosion/spaCy/issues/3052#issuecomment-986756711 and comment https://github.com/explosion/spaCy/issues/3052#issuecomment-986951831
* Bump sudachipy version
* Update sudachipy versions
* Bump versions
Bumping to the most recent dictionary just to keep thing current.
Bumping sudachipy to 5.2 because older versions don't support recent
dictionaries.
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
Co-authored-by: Richard Hudson <richard@explosion.ai>
Co-authored-by: Duygu Altinok <duygu@explosion.ai>
Co-authored-by: Haakon Meland Eriksen <haakon.eriksen@far.no>
This change changes the type of left/right-arc collections from
vector[ArcC] to unordered_map[int, vector[Arc]], so that the arcs are
keyed by the head. This allows us to find all the left/right arcs for a
particular head in constant time in StateC::{L,R}.
Benchmarks with long docs (N is the number of text repetitions):
Before (using #10019):
N Time (s)
400 3.2
800 5.0
1600 9.5
3200 23.2
6400 66.8
12800 220.0
After (this commit):
N Time (s)
400 3.1
800 4.3
1600 6.7
3200 12.0
6400 22.0
12800 42.0
Related to #9858 and #10019.
* Speed up the StateC::L feature function
This function gets the n-th most-recent left-arc with a particular head.
Before this change, StateC::L would construct a vector of all left-arcs
with the given head and then pick the n-th most recent from that vector.
Since the number of left-arcs strongly correlates with the doc length
and the feature is constructed for every transition, this can make
transition-parsing quadratic.
With this change StateC::L:
- Searches left-arcs backwards.
- Stops early when the n-th matching transition is found.
- Does not construct a vector (reducing memory pressure).
This change doesn't avoid the linear search when the transition that is
queried does not occur in the left-arcs. Regardless, performance is
improved quite a bit with very long docs:
Before:
N Time
400 3.3
800 5.4
1600 11.6
3200 30.7
After:
N Time
400 3.2
800 5.0
1600 9.5
3200 23.2
We can probably do better with more tailored data structures, but I
first wanted to make a low-impact PR.
Found while investigating #9858.
* StateC::L: simplify loop
* Speed up the StateC::L feature function
This function gets the n-th most-recent left-arc with a particular head.
Before this change, StateC::L would construct a vector of all left-arcs
with the given head and then pick the n-th most recent from that vector.
Since the number of left-arcs strongly correlates with the doc length
and the feature is constructed for every transition, this can make
transition-parsing quadratic.
With this change StateC::L:
- Searches left-arcs backwards.
- Stops early when the n-th matching transition is found.
- Does not construct a vector (reducing memory pressure).
This change doesn't avoid the linear search when the transition that is
queried does not occur in the left-arcs. Regardless, performance is
improved quite a bit with very long docs:
Before:
N Time
400 3.3
800 5.4
1600 11.6
3200 30.7
After:
N Time
400 3.2
800 5.0
1600 9.5
3200 23.2
We can probably do better with more tailored data structures, but I
first wanted to make a low-impact PR.
Found while investigating #9858.
* StateC::L: simplify loop
