Commit Graph

50 Commits

Author SHA1 Message Date
Adriane Boyd
0fe43f40f1
Support registered vectors (#12492)
* 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>
2023-08-01 15:46:08 +02:00
Madeesh Kannan
5ea14af32b
Add training.before_update callback (#11739)
* Add `training.before_update` callback

This callback can be used to implement training paradigms like gradual (un)freezing of components (e.g: the Transformer) after a certain number of training steps to mitigate catastrophic forgetting during fine-tuning.

* Fix type annotation, default config value

* Generalize arguments passed to the callback

* Update schema

* Pass `epoch` to callback, rename `current_step` to `step`

* Add test

* Simplify test

* Replace config string with `spacy.blank`

* Apply suggestions from code review

Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com>

* Cleanup imports

Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com>
2022-11-23 17:54:58 +01:00
Lj Miranda
00e7bf5ffd
Add a few docs to the default_config.cfg (#9981)
* Clarify patience hyperparameter

The current value for patience doesn't seem to indicate that it's
pointing to the number of steps. It may be useful to specify that
explicitly.

Ref: https://github.com/explosion/spaCy/discussions/7450
Ref: https://github.com/explosion/spaCy/discussions/7465

* Update docs for max_steps
2022-01-05 09:16:40 +01:00
Adriane Boyd
95c0833656
Add training option to set annotations on update (#7767)
* Add training option to set annotations on update

Add a `[training]` option called `set_annotations_on_update` to specify
a list of components for which the predicted annotations should be set
on `example.predicted` immediately after that component has been
updated. The predicted annotations can be accessed by later components
in the pipeline during the processing of the batch in the same `update`
call.

* Rename to annotates / annotating_components

* Add test for `annotating_components` when training from config

* Add documentation
2021-04-26 16:53:53 +02:00
Adriane Boyd
ff84075839
Support large/infinite training corpora (#7208)
* Support infinite generators for training corpora

Support a training corpus with an infinite generator in the `spacy
train` training loop:

* Revert `create_train_batches` to the state where an infinite generator
can be used as the in the first epoch of exactly one epoch without
resulting in a memory leak (`max_epochs != 1` will still result in a
memory leak)
* Move the shuffling for the first epoch into the corpus reader,
renaming it to `spacy.Corpus.v2`.

* Switch to training option for shuffling in memory

Training loop:

* Add option `training.shuffle_train_corpus_in_memory` that controls
whether the corpus is loaded in memory once and shuffled in the training
loop
  * Revert changes to `create_train_batches` and rename to
`create_train_batches_with_shuffling` for use with `spacy.Corpus.v1` and
a corpus that should be loaded in memory
  * Add `create_train_batches_without_shuffling` for a corpus that
should not be shuffled in the training loop: the corpus is merely
batched during training

Corpus readers:

* Restore `spacy.Corpus.v1`
* Add `spacy.ShuffledCorpus.v1` for a corpus shuffled in memory in the
reader instead of the training loop
  * In combination with `shuffle_train_corpus_in_memory = False`, each
epoch could result in a different augmentation

* Refactor create_train_batches, validation

* Rename config setting to `training.shuffle_train_corpus`
* Refactor to use a single `create_train_batches` method with a
`shuffle` option
* Only validate `get_examples` in initialize step if:
  * labels are required
  * labels are not provided

* Switch back to max_epochs=-1 for streaming train corpus

* Use first 100 examples for stream train corpus init

* Always check validate_get_examples in initialize
2021-04-08 18:08:04 +10:00
Paul O'Leary McCann
cdab341a75 Remove mention of -1 for early stopping (fix #7535)
Maybe this used to work differently, but currently a negative patience
just causes immediate termination.
2021-03-23 11:50:35 +09:00
Adriane Boyd
a45d89f09a Add initialize.before_init and after_init callbacks
Add `initialize.before_init` and `initialize.after_init` callbacks to
the config. The `initialize.before_init` callback is a place to
implement one-time tokenizer customizations that are then saved with the
model.
2021-01-12 13:07:44 +01:00
Adriane Boyd
fa8fa474a3 Add nlp.batch_size setting
Add a default `batch_size` setting for `Language.pipe` and
`Language.evaluate` as `nlp.batch_size`.
2020-12-09 09:13:26 +01:00
Ines Montani
5762876dcc Update default config [ci skip] 2020-10-01 22:27:37 +02:00
Matthew Honnibal
5128298964 Add missing augmenter 2020-09-30 20:18:45 +02:00
Ines Montani
fe3f111c37
Merge pull request #6168 from explosion/fix/default-corpus-values 2020-09-30 00:24:02 +02:00
Matthew Honnibal
a2aa1f6882 Disable the OVL augmentation by default 2020-09-29 23:02:40 +02:00
Ines Montani
1aeef3bfbb Make corpus paths default to None and improve errors 2020-09-29 22:33:46 +02:00
Ines Montani
d3c63b7965 Merge branch 'develop' into feature/prepare 2020-09-29 20:53:05 +02:00
Ines Montani
fd594cfb9b Tighten up format 2020-09-29 16:47:55 +02:00
Ines Montani
a5f2cc0509 Tidy up and remove raw text (rehearsal) for now 2020-09-28 12:30:13 +02:00
Ines Montani
1590de11b1 Update config 2020-09-28 12:05:23 +02:00
Matthew Honnibal
9f6ad06452 Upd default config 2020-09-28 12:00:23 +02:00
Ines Montani
e44a7519cd Update CLI and add [initialize] block 2020-09-28 11:56:14 +02:00
Matthew Honnibal
a976da168c
Support data augmentation in Corpus (#6155)
* Support data augmentation in Corpus

* Note initial docs for data augmentation

* Add augmenter to quickstart

* Fix flake8

* Format

* Fix test

* Update spacy/tests/training/test_training.py

* Improve data augmentation arguments

* Update templates

* Move randomization out into caller

* Refactor

* Update spacy/training/augment.py

* Update spacy/tests/training/test_training.py

* Fix augment

* Fix test
2020-09-28 03:03:27 +02:00
Matthew Honnibal
39b178999c Tmp notes 2020-09-27 20:13:38 +02:00
Ines Montani
be56c0994b Add [training.before_to_disk] callback 2020-09-24 12:40:25 +02:00
Ines Montani
e863b3dc14
Merge pull request #6092 from adrianeboyd/bugfix/load-vocab-lookups-2 2020-09-19 12:33:38 +02:00
Sofie Van Landeghem
39872de1f6
Introducing the gpu_allocator (#6091)
* rename 'use_pytorch_for_gpu_memory' to 'gpu_allocator'

* --code instead of --code-path

* update documentation

* avoid querying the "system" section directly

* add explanation of gpu_allocator to TF/PyTorch section in docs

* fix typo

* fix typo 2

* use set_gpu_allocator from thinc 8.0.0a34

* default null instead of empty string
2020-09-19 01:17:02 +02:00
Adriane Boyd
eed4b785f5 Load vocab lookups tables at beginning of training
Similar to how vectors are handled, move the vocab lookups to be loaded
at the start of training rather than when the vocab is initialized,
since the vocab doesn't have access to the full config when it's
created.

The option moves from `nlp.load_vocab_data` to `training.lookups`.

Typically these tables will come from `spacy-lookups-data`, but any
`Lookups` object can be provided.

The loading from `spacy-lookups-data` is now strict, so configs for each
language should specify the exact tables required. This also makes it
easier to control whether the larger clusters and probs tables are
included.

To load `lexeme_norm` from `spacy-lookups-data`:

```
[training.lookups]
@misc = "spacy.LoadLookupsData.v1"
lang = ${nlp.lang}
tables = ["lexeme_norm"]
```
2020-09-18 15:59:16 +02:00
svlandeg
0c35885751 generalize corpora, dot notation for dev and train corpus 2020-09-17 11:38:59 +02:00
svlandeg
51fa929f47 rewrite train_corpus to corpus.train in config 2020-09-15 21:58:04 +02:00
Matthew Honnibal
4b82882767 Fix defaults 2020-09-08 15:31:21 +02:00
Ines Montani
c063e55eb7 Add prefix to batchers 2020-09-03 17:30:41 +02:00
Ines Montani
3ce5be4b76 Allow loaded but disabled components 2020-08-28 15:20:14 +02:00
Sofie Van Landeghem
79d460e3a2
Weights & Biases logger for train CLI (#5971)
* quick test as part of train script

* train_logger in config, default ConsoleLogger in loggers catalogue

* entitiy typo

* add wandb_logger

* cleanup

* Update spacy/cli/train_logger.py

Co-authored-by: Ines Montani <ines@ines.io>

* move loggers to gold.loggers

Co-authored-by: Ines Montani <ines@ines.io>
2020-08-26 15:24:33 +02:00
Ines Montani
0e7f99da58
Fix handling of optional [pretraining] block (#5954)
* Fix handling of optional [pretraining] block

* Remote pretraining from default config

* Fix test

* Add schema option for empty pretrain block
2020-08-24 15:56:03 +02:00
Ines Montani
6ad59d59fe Merge branch 'develop' of https://github.com/explosion/spaCy into develop [ci skip] 2020-08-20 11:20:58 +02:00
Adriane Boyd
e962784531
Add Lemmatizer and simplify related components (#5848)
* Add Lemmatizer and simplify related components

* Add `Lemmatizer` pipe with `lookup` and `rule` modes using the
`Lookups` tables.
* Reduce `Tagger` to a simple tagger that sets `Token.tag` (no pos or lemma)
* Reduce `Morphology` to only keep track of morph tags (no tag map, lemmatizer,
or morph rules)
* Remove lemmatizer from `Vocab`
* Adjust many many tests

Differences:

* No default lookup lemmas
* No special treatment of TAG in `from_array` and similar required
* Easier to modify labels in a `Tagger`
* No extra strings added from morphology / tag map

* Fix test

* Initial fix for Lemmatizer config/serialization

* Adjust init test to be more generic

* Adjust init test to force empty Lookups

* Add simple cache to rule-based lemmatizer

* Convert language-specific lemmatizers

Convert language-specific lemmatizers to component lemmatizers. Remove
previous lemmatizer class.

* Fix French and Polish lemmatizers

* Remove outdated UPOS conversions

* Update Russian lemmatizer init in tests

* Add minimal init/run tests for custom lemmatizers

* Add option to overwrite existing lemmas

* Update mode setting, lookup loading, and caching

* Make `mode` an immutable property
* Only enforce strict `load_lookups` for known supported modes
* Move caching into individual `_lemmatize` methods

* Implement strict when lang is not found in lookups

* Fix tables/lookups in make_lemmatizer

* Reallow provided lookups and allow for stricter checks

* Add lookups asset to all Lemmatizer pipe tests

* Rename lookups in lemmatizer init test

* Clean up merge

* Refactor lookup table loading

* Add helper from `load_lemmatizer_lookups` that loads required and
optional lookups tables based on settings provided by a config.

Additional slight refactor of lookups:

* Add `Lookups.set_table` to set a table from a provided `Table`
* Reorder class definitions to be able to specify type as `Table`

* Move registry assets into test methods

* Refactor lookups tables config

Use class methods within `Lemmatizer` to provide the config for
particular modes and to load the lookups from a config.

* Add pipe and score to lemmatizer

* Simplify Tagger.score

* Add missing import

* Clean up imports and auto-format

* Remove unused kwarg

* Tidy up and auto-format

* Update docstrings for Lemmatizer

Update docstrings for Lemmatizer.

Additionally modify `is_base_form` API to take `Token` instead of
individual features.

* Update docstrings

* Remove tag map values from Tagger.add_label

* Update API docs

* Fix relative link in Lemmatizer API docs
2020-08-07 15:27:13 +02:00
Ines Montani
823e533dc1
Add config callbacks for modifying nlp object before and after init (#5866)
* WIP: Concept for modifying nlp object before and after init

* Make callbacks return nlp object

Co-authored-by: Matthew Honnibal <honnibal+gh@gmail.com>

* Raise if callbacks don't return correct type

* Rename, update types, add after_pipeline_creation

Co-authored-by: Matthew Honnibal <honnibal+gh@gmail.com>
2020-08-05 19:47:54 +02:00
Ines Montani
b795f02fbd
Allow adding pipeline components from source model (#5857)
* Allow adding pipeline components from source model

* Config: name -> component

* Improve error messages

* Fix error and test

* Add frozen components and exclude logic

* Remove exclude from Language.evaluate

* Init sourced components with current vocab

* Fix error codes
2020-08-04 23:39:19 +02:00
Matthew Honnibal
ecb3c4e8f4
Create corpus iterator and batcher from registry during training (#5865)
* Move batchers into their own module (and registry)

* Update CLI

* Update Corpus and batcher

* Update tests

* Update one config

* Merge 'evaluation' block back under [training]

* Import batchers in gold __init__

* Fix batchers

* Update config

* Update schema

* Update util

* Don't assume train and dev are actually paths

* Update onto-joint config

* Fix missing import

* Format

* Format

* Update spacy/gold/corpus.py

Co-authored-by: Ines Montani <ines@ines.io>

* Fix name

* Update default config

* Fix get_length option in batchers

* Update test

* Add comment

* Pass path into Corpus

* Update docstring

* Update schema and configs

* Update config

* Fix test

* Fix paths

* Fix print

* Fix create_train_batches

* [training.read_train] -> [training.train_corpus]

* Update onto-joint config

Co-authored-by: Ines Montani <ines@ines.io>
2020-08-04 15:09:37 +02:00
Ines Montani
0094cb0d04 Remove scores list from config and document 2020-07-28 11:22:24 +02:00
Ines Montani
894e20c466 Merge branch 'develop' into feature/component-scores 2020-07-27 18:14:39 +02:00
Matthew Honnibal
80271ac0ba Update default config 2020-07-26 15:27:39 +02:00
Ines Montani
2470486543 Allow pipeline components to set default scores and weights 2020-07-26 13:18:43 +02:00
Ines Montani
cdbd6ba912
Merge pull request #5798 from explosion/feature/language-data-config 2020-07-25 13:34:49 +02:00
Adriane Boyd
2bcceb80c4
Refactor the Scorer to improve flexibility (#5731)
* Refactor the Scorer to improve flexibility

Refactor the `Scorer` to improve flexibility for arbitrary pipeline
components.

* Individual pipeline components provide their own `evaluate` methods
that score a list of `Example`s and return a dictionary of scores
* `Scorer` is initialized either:
  * with a provided pipeline containing components to be scored
  * with a default pipeline containing the built-in statistical
    components (senter, tagger, morphologizer, parser, ner)
* `Scorer.score` evaluates a list of `Example`s and returns a dictionary
of scores referring to the scores provided by the components in the
pipeline

Significant differences:

* `tags_acc` is renamed to `tag_acc` to be consistent with `token_acc`
and the new `morph_acc`, `pos_acc`, and `lemma_acc`
* Scoring is no longer cumulative: `Scorer.score` scores a list of
examples rather than a single example and does not retain any state
about previously scored examples
* PRF values in the returned scores are no longer multiplied by 100

* Add kwargs to Morphologizer.evaluate

* Create generalized scoring methods in Scorer

* Generalized static scoring methods are added to `Scorer`
  * Methods require an attribute (either on Token or Doc) that is
used to key the returned scores

Naming differences:

* `uas`, `las`, and `las_per_type` in the scores dict are renamed to
`dep_uas`, `dep_las`, and `dep_las_per_type`

Scoring differences:

* `Doc.sents` is now scored as spans rather than on sentence-initial
token positions so that `Doc.sents` and `Doc.ents` can be scored with
the same method (this lowers scores since a single incorrect sentence
start results in two incorrect spans)

* Simplify / extend hasattr check for eval method

* Add hasattr check to tokenizer scoring
* Simplify to hasattr check for component scoring

* Reset Example alignment if docs are set

Reset the Example alignment if either doc is set in case the
tokenization has changed.

* Add PRF tokenization scoring for tokens as spans

Add PRF scores for tokens as character spans. The scores are:

* token_acc: # correct tokens / # gold tokens
* token_p/r/f: PRF for (token.idx, token.idx + len(token))

* Add docstring to Scorer.score_tokenization

* Rename component.evaluate() to component.score()

* Update Scorer API docs

* Update scoring for positive_label in textcat

* Fix TextCategorizer.score kwargs

* Update Language.evaluate docs

* Update score names in default config
2020-07-25 12:53:02 +02:00
Ines Montani
8d9d28eb8b Re-add setting for vocab data and tidy up 2020-07-25 12:14:28 +02:00
Ines Montani
b9aaa4e457 Improve vocab data integration and warning 2020-07-25 11:51:30 +02:00
Ines Montani
38f6ea7a78 Simplify language data and revert detailed configs 2020-07-24 14:50:26 +02:00
Ines Montani
b507f61629 Tidy up and move noun_chunks, token_match, url_match 2020-07-22 22:18:46 +02:00
Ines Montani
0fcd352179 Remove omit_extra_lookups 2020-07-22 16:01:17 +02:00
Ines Montani
945f795a3e WIP: move more language data to config 2020-07-22 15:59:37 +02:00
Ines Montani
43b960c01b
Refactor pipeline components, config and language data (#5759)
* Update with WIP

* Update with WIP

* Update with pipeline serialization

* Update types and pipe factories

* Add deep merge, tidy up and add tests

* Fix pipe creation from config

* Don't validate default configs on load

* Update spacy/language.py

Co-authored-by: Ines Montani <ines@ines.io>

* Adjust factory/component meta error

* Clean up factory args and remove defaults

* Add test for failing empty dict defaults

* Update pipeline handling and methods

* provide KB as registry function instead of as object

* small change in test to make functionality more clear

* update example script for EL configuration

* Fix typo

* Simplify test

* Simplify test

* splitting pipes.pyx into separate files

* moving default configs to each component file

* fix batch_size type

* removing default values from component constructors where possible (TODO: test 4725)

* skip instead of xfail

* Add test for config -> nlp with multiple instances

* pipeline.pipes -> pipeline.pipe

* Tidy up, document, remove kwargs

* small cleanup/generalization for Tok2VecListener

* use DEFAULT_UPSTREAM field

* revert to avoid circular imports

* Fix tests

* Replace deprecated arg

* Make model dirs require config

* fix pickling of keyword-only arguments in constructor

* WIP: clean up and integrate full config

* Add helper to handle function args more reliably

Now also includes keyword-only args

* Fix config composition and serialization

* Improve config debugging and add visual diff

* Remove unused defaults and fix type

* Remove pipeline and factories from meta

* Update spacy/default_config.cfg

Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>

* Update spacy/default_config.cfg

* small UX edits

* avoid printing stack trace for debug CLI commands

* Add support for language-specific factories

* specify the section of the config which holds the model to debug

* WIP: add Language.from_config

* Update with language data refactor WIP

* Auto-format

* Add backwards-compat handling for Language.factories

* Update morphologizer.pyx

* Fix morphologizer

* Update and simplify lemmatizers

* Fix Japanese tests

* Port over tagger changes

* Fix Chinese and tests

* Update to latest Thinc

* WIP: xfail first Russian lemmatizer test

* Fix component-specific overrides

* fix nO for output layers in debug_model

* Fix default value

* Fix tests and don't pass objects in config

* Fix deep merging

* Fix lemma lookup data registry

Only load the lookups if an entry is available in the registry (and if spacy-lookups-data is installed)

* Add types

* Add Vocab.from_config

* Fix typo

* Fix tests

* Make config copying more elegant

* Fix pipe analysis

* Fix lemmatizers and is_base_form

* WIP: move language defaults to config

* Fix morphology type

* Fix vocab

* Remove comment

* Update to latest Thinc

* Add morph rules to config

* Tidy up

* Remove set_morphology option from tagger factory

* Hack use_gpu

* Move [pipeline] to top-level block and make [nlp.pipeline] list

Allows separating component blocks from component order – otherwise, ordering the config would mean a changed component order, which is bad. Also allows initial config to define more components and not use all of them

* Fix use_gpu and resume in CLI

* Auto-format

* Remove resume from config

* Fix formatting and error

* [pipeline] -> [components]

* Fix types

* Fix tagger test: requires set_morphology?

Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com>
Co-authored-by: svlandeg <sofie.vanlandeghem@gmail.com>
Co-authored-by: Matthew Honnibal <honnibal+gh@gmail.com>
2020-07-22 13:42:59 +02:00