Before this change, the workers of pipe call with n_process != 1 were
stopped by calling `terminate` on the processes. However, terminating a
process can leave queues, pipes, and other concurrent data structures in
an invalid state.
With this change, we stop using terminate and take the following approach
instead:
* When the all documents are processed, the parent process puts a
sentinel in the queue of each worker.
* The parent process then calls `join` on each worker process to
let them finish up gracefully.
* Worker processes break from the queue processing loop when the
sentinel is encountered, so that they exit.
We need special handling when one of the workers encounters an error and
the error handler is set to raise an exception. In this case, we cannot
rely on the sentinel to finish all workers -- the queue is a FIFO queue
and there may be other work queued up before the sentinel. We use the
following approach to handle error scenarios:
* The parent puts the end-of-work sentinel in the queue of each worker.
* The parent closes the reading-end of the channel of each worker.
* Then:
- If the worker was waiting for work, it will encounter the sentinel
and break from the processing loop.
- If the worker was processing a batch, it will attempt to write
results to the channel. This will fail because the channel was
closed by the parent and the worker will break from the processing
loop.
* Support registered vectors
* Format
* Auto-fill [nlp] on load from config and from bytes/disk
* Only auto-fill [nlp]
* Undo all changes to Language.from_disk
* Expand BaseVectors
These methods are needed in various places for training and vector
similarity.
* isort
* More linting
* Only fill [nlp.vectors]
* Update spacy/vocab.pyx
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
* Revert changes to test related to auto-filling [nlp]
* Add vectors registry
* Rephrase error about vocab methods for vectors
* Switch to dummy implementation for BaseVectors.to_ops
* Add initial draft of docs
* Remove example from BaseVectors docs
* Apply suggestions from code review
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
* Update website/docs/api/basevectors.mdx
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
* Fix type and lint bpemb example
* Update website/docs/api/basevectors.mdx
---------
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
* Recommend lookups tables from URLs or other loaders
Shift away from the `lookups` extra (which isn't removed, just no longer
mentioned) and recommend loading data from the `spacy-lookups-data` repo
or other sources rather than the `spacy-lookups-data` package.
If the tables can't be loaded from the `lookups` registry in the
lemmatizer, show how to specify the tables in `[initialize]` rather than
recommending the `spacy-lookups-data` package.
* Add tests for some rule-based lemmatizers
* Apply suggestions from code review
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
---------
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
* modified: spacy/language.py
- corrected typo in docstring for :method:`Language.replace_listeners`
- added noqa comment on unused local variable assignment in :method:`Language.from_config` as I wasn't sure if it should be unassigned
modified: website/docs/api/language.mdx
- corrected typo in `Language.replace_listeners` markdown
* modified: spacy/language.py
- removed noqa comment
---------
Co-authored-by: Ian Thompson <ian.thompson@hrblock.com>
* Literal True for first/last options
* add test case
* update docs
* remove old redundant test case
* black formatting
* use Optional typing in docstrings
Co-authored-by: Raphael Mitsch <r.mitsch@outlook.com>
---------
Co-authored-by: Raphael Mitsch <r.mitsch@outlook.com>
* `Language.replace_listeners`: Pass the replaced listener and the `tok2vec` pipe to the callback
* Update developer docs
* `isort` fixes
* Add error message to assertion
* Add clarification to dev docs
* Replace assertion with exception
* Doc fixes
When sourcing a component, the object from the original pipeline is added to the new pipeline as the same object. This creates a situation where there are several attributes that cannot be in sync between the original pipeline and the new pipeline at the same time for this one object:
* component.name
* component.listener_map / component.listening_components for tok2vec and transformer
When running replace_listeners on a component, the config is not updated correctly if the state of the component is incorrect for the current pipeline (in particular changes that should be applied from model.attrs["replace_listener_cfg"] as used in spacy-transformers) due to the fact that:
* find_listeners relies on component.name to set the name in the listener_map
* replace_listeners relies on listener_map to determine how to modify the configs
In addition, there are several places where pipeline components are modified and the listener map and/or internal component names aren't currently updated.
In cases where there is a component shared by two pipelines that cannot be in sync, this PR chooses to prioritize the most recently modified or initialized pipeline. There is no actual solution with the current source behavior that will make both pipelines usable, so the current pipeline is updated whenever components are added/renamed/removed or the pipeline is initialized for training.
* Use isort with Black profile
* isort all the things
* Fix import cycles as a result of import sorting
* Add DOCBIN_ALL_ATTRS type definition
* Add isort to requirements
* Remove isort from build dependencies check
* Typo
* Add scorer option to return per-component scores
Add `per_component` option to `Language.evaluate` and `Scorer.score` to
return scores keyed by `tokenizer` (hard-coded) or by component name.
Add option to `evaluate` CLI to score by component. Per-component scores
can only be saved to JSON.
* Update help text and messages
* Add distillation initialization and loop
* Fix up configuration keys
* Add docstring
* Type annotations
* init_nlp_distill -> init_nlp_student
* Do not resolve dot name distill corpus in initialization
(Since we don't use it.)
* student: do not request use of optimizer in student pipe
We apply finish up the updates once in the training loop instead.
Also add the necessary logic to `Language.distill` to mirror
`Language.update`.
* Correctly determine sort key in subdivide_batch
* Fix _distill_loop docstring wrt. stopping condition
* _distill_loop: fix distill_data docstring
Make similar changes in train_while_improving, since it also had
incorrect types and missing type annotations.
* Move `set_{gpu_allocator,seed}_from_config` to spacy.util
* Update Language.update docs for the sgd argument
* Type annotation
Co-authored-by: Madeesh Kannan <shadeMe@users.noreply.github.com>
---------
Co-authored-by: Madeesh Kannan <shadeMe@users.noreply.github.com>
* Avoid `TrainablePipe.finish_update` getting called twice during training
PR #12136 fixed an issue where the tok2vec pipe was updated before
gradient were accumulated. However, it introduced a new bug that cause
`finish_update` to be called twice when using the training loop. This
causes a fairly large slowdown.
The `Language.update` method accepts the `sgd` argument for passing an
optimizer. This argument has three possible values:
- `Optimizer`: use the given optimizer to finish pipe updates.
- `None`: use a default optimizer to finish pipe updates.
- `False`: do not finish pipe updates.
However, the latter option was not documented and not valid with the
existing type of `sgd`. I assumed that this was a remnant of earlier
spaCy versions and removed handling of `False`.
However, with that change, we are passing `None` to `Language.update`.
As a result, we were calling `finish_update` in both `Language.update`
and in the training loop after all subbatches are processed.
This change restores proper handling/use of `False`. Moreover, the role
of `False` is now documented and added to the type to avoid future
accidents.
* Fix typo
* Document defaults for `Language.update`
* `Language.update`: ensure that tok2vec gets updated
The components in a pipeline can be updated independently. However,
tok2vec implementations are an exception to this, since they depend on
listeners for their gradients. The update method of a tok2vec
implementation computes the tok2vec forward and passes this along with a
backprop function to the listeners. This backprop function accumulates
gradients for all the listeners. There are two ways in which the
accumulated gradients can be used to update the tok2vec weights:
1. Call the `finish_update` method of tok2vec *after* the `update`
method is called on all of the pipes that use a tok2vec listener.
2. Pass an optimizer to the `update` method of tok2vec. In this
case, tok2vec will give the last listener a special backprop
function that calls `finish_update` on the tok2vec.
Unfortunately, `Language.update` did neither of these. Instead, it
immediately called `finish_update` on every pipe after `update`. As a
result, the tok2vec weights are updated when no gradients have been
accumulated from listeners yet. And the gradients of the listeners are
only used in the next call to `Language.update` (when `finish_update` is
called on tok2vec again).
This change fixes this issue by passing the optimizer to the `update`
method of trainable pipes, leading to use of the second strategy
outlined above.
The main updating loop in `Language.update` is also simplified by using
the `TrainableComponent` protocol consistently.
* Train loop: `sgd` is `Optional[Optimizer]`, do not pass false
* Language.update: call pipe finish_update after all pipe updates
This does correct and fast updates if multiple components update the
same parameters.
* Add comment why we moved `finish_update` to a separate loop
* change logging call for spacy.LookupsDataLoader.v1
* substitutions in language and _util
* various more substitutions
* add string formatting guidelines to contribution guidelines
* Language.distill: copy both reference and predicted
In distillation we also modify the teacher docs (e.g. in tok2vec
components), so we need to copy both the reference and predicted doc.
Problem caught by @shadeMe
* Make new `_copy_examples` args kwonly
* Add the configuration schema for distillation
This also adds the default configuration and some tests. The schema will
be used by the training loop and `distill` subcommand.
* Format
* Change distillation shortopt to -d
* Fix descripion of max_epochs
* Rename distillation flag to -dt
* Rename `pipe_map` to `student_to_teacher`
* Add `Language.distill`
This method is the distillation counterpart of `Language.update`. It
takes a teacher `Language` instance and distills the student pipes on
the teacher pipes.
* Apply suggestions from code review
Co-authored-by: Madeesh Kannan <shadeMe@users.noreply.github.com>
* Clarify that how Example is used in distillation
* Update transition parser distill docstring for examples argument
* Pass optimizer to `TrainablePipe.distill`
* Annotate pipe before update
As discussed internally, we want to let a pipe annotate before doing an
update with gold/silver data. Otherwise, the output may be (too)
informed by the gold/silver data.
* Rename `component_map` to `student_to_teacher`
* Better synopsis in `Language.distill` docstring
* `name` -> `student_name`
* Fix labels type in docstring
* Mark distill test as slow
* Fix `student_to_teacher` type in docs
---------
Co-authored-by: Madeesh Kannan <shadeMe@users.noreply.github.com>
* Add equality definition for vectors
This re-uses the check from sourcing components.
* Use the equality check
* Format
Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com>
* Change enable/disable behavior so that arguments take precedence over config options. Extend error message on conflict. Add warning message in case of overwriting config option with arguments.
* Fix tests in test_serialize_pipeline.py to reflect changes to handling of enable/disable.
* Fix type issue.
* Move comment.
* Move comment.
* Issue UserWarning instead of printing wasabi message. Adjust test.
* Added pytest.warns(UserWarning) for expected warning to fix tests.
* Update warning message.
* Move type handling out of fetch_pipes_status().
* Add global variable for default value. Use id() to determine whether used values are default value.
* Fix default value for disable.
* Rename DEFAULT_PIPE_STATUS to _DEFAULT_EMPTY_PIPES.
* adding unit test for spacy.load with disable/exclude string arg
* allow pure strings in from_config
* update docs
* upstream type adjustements
* docs update
* make docstring more consistent
* Update spacy/language.py
Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com>
* two more cleanups
* fix type in internal method
Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com>
* Enable flag on spacy.load: foundation for include, enable arguments.
* Enable flag on spacy.load: fixed tests.
* Enable flag on spacy.load: switched from pretrained model to empty model with added pipes for tests.
* Enable flag on spacy.load: switched to more consistent error on misspecification of component activity. Test refactoring. Added to default config.
* Enable flag on spacy.load: added support for fields not in pipeline.
* Enable flag on spacy.load: removed serialization fields from supported fields.
* Enable flag on spacy.load: removed 'enable' from config again.
* Enable flag on spacy.load: relaxed checks in _resolve_component_activation_status() to allow non-standard pipes.
* Enable flag on spacy.load: fixed relaxed checks for _resolve_component_activation_status() to allow non-standard pipes. Extended tests.
* Enable flag on spacy.load: comments w.r.t. resolution workarounds.
* Enable flag on spacy.load: remove include fields. Update website docs.
* Enable flag on spacy.load: updates w.r.t. changes in master.
* Implement Doc.from_json(): update docstrings.
Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com>
* Implement Doc.from_json(): remove newline.
Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com>
* Implement Doc.from_json(): change error message for E1038.
Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com>
* Enable flag on spacy.load: wrapped docstring for _resolve_component_status() at 80 chars.
* Enable flag on spacy.load: changed exmples for enable flag.
* Remove newline.
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
* Fix docstring for Language._resolve_component_status().
* Rename E1038 to E1042.
Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com>
Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
* Pipe name override in config: added check with warning, added removal of name override from config, extended tests.
* Pipoe name override in config: added pytest UserWarning.
Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com>
Co-authored-by: Adriane Boyd <adrianeboyd@gmail.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>
* 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>
* 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>
* 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
Exclude strings from `Vector.to_bytes()` comparions for v3.2+ `Vectors`
that now include the string store so that the source vector comparison
is only comparing the vectors and not the strings.
* make nlp.pipe() return None docs when no exceptions are (re-)raised during error handling
* Remove changes other than as_tuples test
* Only check warning count for one process
* Fix types
* Format
Co-authored-by: Xi Bai <xi.bai.ed@gmail.com>
* Add support for fasttext-bloom hash-only vectors
Overview:
* Extend `Vectors` to have two modes: `default` and `ngram`
* `default` is the default mode and equivalent to the current
`Vectors`
* `ngram` supports the hash-only ngram tables from `fasttext-bloom`
* Extend `spacy.StaticVectors.v2` to handle both modes with no changes
for `default` vectors
* Extend `spacy init vectors` to support ngram tables
The `ngram` mode **only** supports vector tables produced by this
fork of fastText, which adds an option to represent all vectors using
only the ngram buckets table and which uses the exact same ngram
generation algorithm and hash function (`MurmurHash3_x64_128`).
`fasttext-bloom` produces an additional `.hashvec` table, which can be
loaded by `spacy init vectors --fasttext-bloom-vectors`.
https://github.com/adrianeboyd/fastText/tree/feature/bloom
Implementation details:
* `Vectors` now includes the `StringStore` as `Vectors.strings` so that
the API can stay consistent for both `default` (which can look up from
`str` or `int`) and `ngram` (which requires `str` to calculate the
ngrams).
* In ngram mode `Vectors` uses a default `Vectors` object as a cache
since the ngram vectors lookups are relatively expensive.
* The default cache size is the same size as the provided ngram vector
table.
* Once the cache is full, no more entries are added. The user is
responsible for managing the cache in cases where the initial
documents are not representative of the texts.
* The cache can be resized by setting `Vectors.ngram_cache_size` or
cleared with `vectors._ngram_cache.clear()`.
* The API ends up a bit split between methods for `default` and for
`ngram`, so functions that only make sense for `default` or `ngram`
include warnings with custom messages suggesting alternatives where
possible.
* `Vocab.vectors` becomes a property so that the string stores can be
synced when assigning vectors to a vocab.
* `Vectors` serializes its own config settings as `vectors.cfg`.
* The `Vectors` serialization methods have added support for `exclude`
so that the `Vocab` can exclude the `Vectors` strings while serializing.
Removed:
* The `minn` and `maxn` options and related code from
`Vocab.get_vector`, which does not work in a meaningful way for default
vector tables.
* The unused `GlobalRegistry` in `Vectors`.
* Refactor to use reduce_mean
Refactor to use reduce_mean and remove the ngram vectors cache.
* Rename to floret
* Rename to floret in error messages
* Use --vectors-mode in CLI, vector init
* Fix vectors mode in init
* Remove unused var
* Minor API and docstrings adjustments
* Rename `--vectors-mode` to `--mode` in `init vectors` CLI
* Rename `Vectors.get_floret_vectors` to `Vectors.get_batch` and support
both modes.
* Minor updates to Vectors docstrings.
* Update API docs for Vectors and init vectors CLI
* Update types for StaticVectors
