spaCy/spacy/training/initialize.py

363 lines
14 KiB
Python
Raw Permalink Normal View History

2020-09-29 08:58:50 +00:00
import gzip
import tarfile
import warnings
2020-09-29 08:58:50 +00:00
import zipfile
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 08:08:04 +00:00
from itertools import islice
from pathlib import Path
from typing import IO, TYPE_CHECKING, Any, Dict, Optional, Union
import numpy
import srsly
import tqdm
from thinc.api import Config, ConfigValidationError, fix_random_seed, set_gpu_allocator
2020-09-28 13:09:59 +00:00
from ..errors import Errors, Warnings
from ..lookups import Lookups
from ..schemas import ConfigSchemaTraining
from ..util import (
DEFAULT_OOV_PROB,
OOV_RANK,
ensure_path,
get_sourced_components,
load_model,
load_model_from_config,
logger,
registry,
resolve_dot_names,
)
from ..vectors import Mode as VectorsMode
from ..vectors import Vectors
from .pretrain import get_tok2vec_ref
2020-09-28 13:09:59 +00:00
if TYPE_CHECKING:
from ..language import Language # noqa: F401
2020-09-28 13:09:59 +00:00
2020-09-29 14:08:39 +00:00
def init_nlp(config: Config, *, use_gpu: int = -1) -> "Language":
2020-09-28 13:09:59 +00:00
raw_config = config
config = raw_config.interpolate()
if "seed" not in config["training"]:
raise ValueError(Errors.E1015.format(value="[training] seed"))
if "gpu_allocator" not in config["training"]:
raise ValueError(Errors.E1015.format(value="[training] gpu_allocator"))
2020-09-28 13:09:59 +00:00
if config["training"]["seed"] is not None:
fix_random_seed(config["training"]["seed"])
allocator = config["training"]["gpu_allocator"]
if use_gpu >= 0 and allocator:
set_gpu_allocator(allocator)
# Use original config here before it's resolved to functions
sourced = get_sourced_components(config)
2020-09-28 13:09:59 +00:00
nlp = load_model_from_config(raw_config, auto_fill=True)
2020-09-29 14:08:39 +00:00
logger.info("Set up nlp object from config")
2020-09-28 13:09:59 +00:00
config = nlp.config.interpolate()
# Resolve all training-relevant sections using the filled nlp config
T = registry.resolve(config["training"], schema=ConfigSchemaTraining)
dot_names = [T["train_corpus"], T["dev_corpus"]]
if not isinstance(T["train_corpus"], str):
2021-01-05 02:41:53 +00:00
raise ConfigValidationError(
desc=Errors.E897.format(
field="training.train_corpus", type=type(T["train_corpus"])
)
)
if not isinstance(T["dev_corpus"], str):
2021-01-05 02:41:53 +00:00
raise ConfigValidationError(
desc=Errors.E897.format(
field="training.dev_corpus", type=type(T["dev_corpus"])
)
)
2020-09-28 13:09:59 +00:00
train_corpus, dev_corpus = resolve_dot_names(config, dot_names)
optimizer = T["optimizer"]
# Components that shouldn't be updated during training
frozen_components = T["frozen_components"]
# Sourced components that require resume_training
resume_components = [p for p in sourced if p not in frozen_components]
logger.info("Pipeline: %s", nlp.pipe_names)
2020-09-28 13:09:59 +00:00
if resume_components:
with nlp.select_pipes(enable=resume_components):
logger.info("Resuming training for: %s", resume_components)
2020-09-28 13:09:59 +00:00
nlp.resume_training(sgd=optimizer)
# Make sure that internal component names are synced and listeners are
# defined before initializing further
nlp._link_components()
2020-09-28 13:09:59 +00:00
with nlp.select_pipes(disable=[*frozen_components, *resume_components]):
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 08:08:04 +00:00
if T["max_epochs"] == -1:
2021-07-17 03:43:15 +00:00
sample_size = 100
2021-06-28 09:48:00 +00:00
logger.debug(
"Due to streamed train corpus, using only first %s examples for initialization. "
"If necessary, provide all labels in [initialize]. "
"More info: https://spacy.io/api/cli#init_labels",
sample_size,
2021-06-28 09:48:00 +00:00
)
2021-07-23 08:04:09 +00:00
nlp.initialize(
lambda: islice(train_corpus(nlp), sample_size), sgd=optimizer
)
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 08:08:04 +00:00
else:
nlp.initialize(lambda: train_corpus(nlp), sgd=optimizer)
logger.info("Initialized pipeline components: %s", nlp.pipe_names)
# Detect components with listeners that are not frozen consistently
for name, proc in nlp.pipeline:
2021-06-28 09:48:00 +00:00
for listener in getattr(
proc, "listening_components", []
): # e.g. tok2vec/transformer
# Don't warn about components not in the pipeline
if listener not in nlp.pipe_names:
continue
if listener in frozen_components and name not in frozen_components:
logger.warning(Warnings.W087.format(name=name, listener=listener))
# We always check this regardless, in case user freezes tok2vec
if listener not in frozen_components and name in frozen_components:
if name not in T["annotating_components"]:
logger.warning(Warnings.W086.format(name=name, listener=listener))
2020-09-28 13:09:59 +00:00
return nlp
def init_vocab(
nlp: "Language",
2020-09-28 13:09:59 +00:00
*,
data: Optional[Path] = None,
lookups: Optional[Lookups] = None,
vectors: Optional[str] = None,
🏷 Add Mypy check to CI and ignore all existing Mypy errors (#9167) * 🚨 Ignore all existing Mypy errors * 🏗 Add Mypy check to CI * Add types-mock and types-requests as dev requirements * Add additional type ignore directives * Add types packages to dev-only list in reqs test * Add types-dataclasses for python 3.6 * Add ignore to pretrain * 🏷 Improve type annotation on `run_command` helper The `run_command` helper previously declared that it returned an `Optional[subprocess.CompletedProcess]`, but it isn't actually possible for the function to return `None`. These changes modify the type annotation of the `run_command` helper and remove all now-unnecessary `# type: ignore` directives. * 🔧 Allow variable type redefinition in limited contexts These changes modify how Mypy is configured to allow variables to have their type automatically redefined under certain conditions. The Mypy documentation contains the following example: ```python def process(items: List[str]) -> None: # 'items' has type List[str] items = [item.split() for item in items] # 'items' now has type List[List[str]] ... ``` This configuration change is especially helpful in reducing the number of `# type: ignore` directives needed to handle the common pattern of: * Accepting a filepath as a string * Overwriting the variable using `filepath = ensure_path(filepath)` These changes enable redefinition and remove all `# type: ignore` directives rendered redundant by this change. * 🏷 Add type annotation to converters mapping * 🚨 Fix Mypy error in convert CLI argument verification * 🏷 Improve type annotation on `resolve_dot_names` helper * 🏷 Add type annotations for `Vocab` attributes `strings` and `vectors` * 🏷 Add type annotations for more `Vocab` attributes * 🏷 Add loose type annotation for gold data compilation * 🏷 Improve `_format_labels` type annotation * 🏷 Fix `get_lang_class` type annotation * 🏷 Loosen return type of `Language.evaluate` * 🏷 Don't accept `Scorer` in `handle_scores_per_type` * 🏷 Add `string_to_list` overloads * 🏷 Fix non-Optional command-line options * 🙈 Ignore redefinition of `wandb_logger` in `loggers.py` * ➕ Install `typing_extensions` in Python 3.8+ The `typing_extensions` package states that it should be used when "writing code that must be compatible with multiple Python versions". Since SpaCy needs to support multiple Python versions, it should be used when newer `typing` module members are required. One example of this is `Literal`, which is available starting with Python 3.8. Previously SpaCy tried to import `Literal` from `typing`, falling back to `typing_extensions` if the import failed. However, Mypy doesn't seem to be able to understand what `Literal` means when the initial import means. Therefore, these changes modify how `compat` imports `Literal` by always importing it from `typing_extensions`. These changes also modify how `typing_extensions` is installed, so that it is a requirement for all Python versions, including those greater than or equal to 3.8. * 🏷 Improve type annotation for `Language.pipe` These changes add a missing overload variant to the type signature of `Language.pipe`. Additionally, the type signature is enhanced to allow type checkers to differentiate between the two overload variants based on the `as_tuple` parameter. Fixes #8772 * ➖ Don't install `typing-extensions` in Python 3.8+ After more detailed analysis of how to implement Python version-specific type annotations using SpaCy, it has been determined that by branching on a comparison against `sys.version_info` can be statically analyzed by Mypy well enough to enable us to conditionally use `typing_extensions.Literal`. This means that we no longer need to install `typing_extensions` for Python versions greater than or equal to 3.8! 🎉 These changes revert previous changes installing `typing-extensions` regardless of Python version and modify how we import the `Literal` type to ensure that Mypy treats it properly. * resolve mypy errors for Strict pydantic types * refactor code to avoid missing return statement * fix types of convert CLI command * avoid list-set confustion in debug_data * fix typo and formatting * small fixes to avoid type ignores * fix types in profile CLI command and make it more efficient * type fixes in projects CLI * put one ignore back * type fixes for render * fix render types - the sequel * fix BaseDefault in language definitions * fix type of noun_chunks iterator - yields tuple instead of span * fix types in language-specific modules * 🏷 Expand accepted inputs of `get_string_id` `get_string_id` accepts either a string (in which case it returns its ID) or an ID (in which case it immediately returns the ID). These changes extend the type annotation of `get_string_id` to indicate that it can accept either strings or IDs. * 🏷 Handle override types in `combine_score_weights` The `combine_score_weights` function allows users to pass an `overrides` mapping to override data extracted from the `weights` argument. Since it allows `Optional` dictionary values, the return value may also include `Optional` dictionary values. These changes update the type annotations for `combine_score_weights` to reflect this fact. * 🏷 Fix tokenizer serialization method signatures in `DummyTokenizer` * 🏷 Fix redefinition of `wandb_logger` These changes fix the redefinition of `wandb_logger` by giving a separate name to each `WandbLogger` version. For backwards-compatibility, `spacy.train` still exports `wandb_logger_v3` as `wandb_logger` for now. * more fixes for typing in language * type fixes in model definitions * 🏷 Annotate `_RandomWords.probs` as `NDArray` * 🏷 Annotate `tok2vec` layers to help Mypy * 🐛 Fix `_RandomWords.probs` type annotations for Python 3.6 Also remove an import that I forgot to move to the top of the module 😅 * more fixes for matchers and other pipeline components * quick fix for entity linker * fixing types for spancat, textcat, etc * bugfix for tok2vec * type annotations for scorer * add runtime_checkable for Protocol * type and import fixes in tests * mypy fixes for training utilities * few fixes in util * fix import * 🐵 Remove unused `# type: ignore` directives * 🏷 Annotate `Language._components` * 🏷 Annotate `spacy.pipeline.Pipe` * add doc as property to span.pyi * small fixes and cleanup * explicit type annotations instead of via comment Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com> Co-authored-by: svlandeg <sofie.vanlandeghem@gmail.com> Co-authored-by: svlandeg <svlandeg@github.com>
2021-10-14 13:21:40 +00:00
) -> None:
2020-09-28 13:09:59 +00:00
if lookups:
nlp.vocab.lookups = lookups
logger.info("Added vocab lookups: %s", ", ".join(lookups.tables))
2020-09-28 13:09:59 +00:00
data_path = ensure_path(data)
if data_path is not None:
lex_attrs = srsly.read_jsonl(data_path)
for lexeme in nlp.vocab:
lexeme.rank = OOV_RANK
for attrs in lex_attrs:
if "settings" in attrs:
continue
lexeme = nlp.vocab[attrs["orth"]]
lexeme.set_attrs(**attrs)
if len(nlp.vocab):
oov_prob = min(lex.prob for lex in nlp.vocab) - 1
else:
oov_prob = DEFAULT_OOV_PROB
nlp.vocab.cfg.update({"oov_prob": oov_prob})
logger.info("Added %d lexical entries to the vocab", len(nlp.vocab))
2020-09-29 14:22:41 +00:00
logger.info("Created vocabulary")
2020-09-28 13:09:59 +00:00
if vectors is not None:
load_vectors_into_model(nlp, vectors)
logger.info("Added vectors: %s", vectors)
# warn if source model vectors are not identical
sourced_vectors_hashes = nlp.meta.pop("_sourced_vectors_hashes", {})
if len(sourced_vectors_hashes) > 0:
vectors_hash = hash(nlp.vocab.vectors.to_bytes(exclude=["strings"]))
for sourced_component, sourced_vectors_hash in sourced_vectors_hashes.items():
if vectors_hash != sourced_vectors_hash:
warnings.warn(Warnings.W113.format(name=sourced_component))
2020-09-29 20:53:14 +00:00
logger.info("Finished initializing nlp object")
2020-09-28 13:09:59 +00:00
def load_vectors_into_model(
nlp: "Language", name: Union[str, Path], *, add_strings: bool = True
2020-09-28 13:09:59 +00:00
) -> None:
"""Load word vectors from an installed model or path into a model instance."""
try:
# Load with the same vocab, which automatically adds the vectors to
# the current nlp object. Exclude lookups so they are not modified.
exclude = ["lookups"]
if not add_strings:
exclude.append("strings")
vectors_nlp = load_model(name, vocab=nlp.vocab, exclude=exclude)
2020-09-28 13:09:59 +00:00
except ConfigValidationError as e:
title = f"Config validation error for vectors {name}"
desc = (
"This typically means that there's a problem in the config.cfg included "
"with the packaged vectors. Make sure that the vectors package you're "
"loading is compatible with the current version of spaCy."
)
err = ConfigValidationError.from_error(e, title=title, desc=desc)
2020-09-28 13:09:59 +00:00
raise err from None
Add support for floret vectors (#8909) * 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
2021-10-27 12:08:31 +00:00
if (
len(vectors_nlp.vocab.vectors.keys()) == 0
and vectors_nlp.vocab.vectors.mode != VectorsMode.floret
) or (
vectors_nlp.vocab.vectors.shape[0] == 0
Add support for floret vectors (#8909) * 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
2021-10-27 12:08:31 +00:00
and vectors_nlp.vocab.vectors.mode == VectorsMode.floret
):
logger.warning(Warnings.W112.format(name=name))
for lex in nlp.vocab:
🏷 Add Mypy check to CI and ignore all existing Mypy errors (#9167) * 🚨 Ignore all existing Mypy errors * 🏗 Add Mypy check to CI * Add types-mock and types-requests as dev requirements * Add additional type ignore directives * Add types packages to dev-only list in reqs test * Add types-dataclasses for python 3.6 * Add ignore to pretrain * 🏷 Improve type annotation on `run_command` helper The `run_command` helper previously declared that it returned an `Optional[subprocess.CompletedProcess]`, but it isn't actually possible for the function to return `None`. These changes modify the type annotation of the `run_command` helper and remove all now-unnecessary `# type: ignore` directives. * 🔧 Allow variable type redefinition in limited contexts These changes modify how Mypy is configured to allow variables to have their type automatically redefined under certain conditions. The Mypy documentation contains the following example: ```python def process(items: List[str]) -> None: # 'items' has type List[str] items = [item.split() for item in items] # 'items' now has type List[List[str]] ... ``` This configuration change is especially helpful in reducing the number of `# type: ignore` directives needed to handle the common pattern of: * Accepting a filepath as a string * Overwriting the variable using `filepath = ensure_path(filepath)` These changes enable redefinition and remove all `# type: ignore` directives rendered redundant by this change. * 🏷 Add type annotation to converters mapping * 🚨 Fix Mypy error in convert CLI argument verification * 🏷 Improve type annotation on `resolve_dot_names` helper * 🏷 Add type annotations for `Vocab` attributes `strings` and `vectors` * 🏷 Add type annotations for more `Vocab` attributes * 🏷 Add loose type annotation for gold data compilation * 🏷 Improve `_format_labels` type annotation * 🏷 Fix `get_lang_class` type annotation * 🏷 Loosen return type of `Language.evaluate` * 🏷 Don't accept `Scorer` in `handle_scores_per_type` * 🏷 Add `string_to_list` overloads * 🏷 Fix non-Optional command-line options * 🙈 Ignore redefinition of `wandb_logger` in `loggers.py` * ➕ Install `typing_extensions` in Python 3.8+ The `typing_extensions` package states that it should be used when "writing code that must be compatible with multiple Python versions". Since SpaCy needs to support multiple Python versions, it should be used when newer `typing` module members are required. One example of this is `Literal`, which is available starting with Python 3.8. Previously SpaCy tried to import `Literal` from `typing`, falling back to `typing_extensions` if the import failed. However, Mypy doesn't seem to be able to understand what `Literal` means when the initial import means. Therefore, these changes modify how `compat` imports `Literal` by always importing it from `typing_extensions`. These changes also modify how `typing_extensions` is installed, so that it is a requirement for all Python versions, including those greater than or equal to 3.8. * 🏷 Improve type annotation for `Language.pipe` These changes add a missing overload variant to the type signature of `Language.pipe`. Additionally, the type signature is enhanced to allow type checkers to differentiate between the two overload variants based on the `as_tuple` parameter. Fixes #8772 * ➖ Don't install `typing-extensions` in Python 3.8+ After more detailed analysis of how to implement Python version-specific type annotations using SpaCy, it has been determined that by branching on a comparison against `sys.version_info` can be statically analyzed by Mypy well enough to enable us to conditionally use `typing_extensions.Literal`. This means that we no longer need to install `typing_extensions` for Python versions greater than or equal to 3.8! 🎉 These changes revert previous changes installing `typing-extensions` regardless of Python version and modify how we import the `Literal` type to ensure that Mypy treats it properly. * resolve mypy errors for Strict pydantic types * refactor code to avoid missing return statement * fix types of convert CLI command * avoid list-set confustion in debug_data * fix typo and formatting * small fixes to avoid type ignores * fix types in profile CLI command and make it more efficient * type fixes in projects CLI * put one ignore back * type fixes for render * fix render types - the sequel * fix BaseDefault in language definitions * fix type of noun_chunks iterator - yields tuple instead of span * fix types in language-specific modules * 🏷 Expand accepted inputs of `get_string_id` `get_string_id` accepts either a string (in which case it returns its ID) or an ID (in which case it immediately returns the ID). These changes extend the type annotation of `get_string_id` to indicate that it can accept either strings or IDs. * 🏷 Handle override types in `combine_score_weights` The `combine_score_weights` function allows users to pass an `overrides` mapping to override data extracted from the `weights` argument. Since it allows `Optional` dictionary values, the return value may also include `Optional` dictionary values. These changes update the type annotations for `combine_score_weights` to reflect this fact. * 🏷 Fix tokenizer serialization method signatures in `DummyTokenizer` * 🏷 Fix redefinition of `wandb_logger` These changes fix the redefinition of `wandb_logger` by giving a separate name to each `WandbLogger` version. For backwards-compatibility, `spacy.train` still exports `wandb_logger_v3` as `wandb_logger` for now. * more fixes for typing in language * type fixes in model definitions * 🏷 Annotate `_RandomWords.probs` as `NDArray` * 🏷 Annotate `tok2vec` layers to help Mypy * 🐛 Fix `_RandomWords.probs` type annotations for Python 3.6 Also remove an import that I forgot to move to the top of the module 😅 * more fixes for matchers and other pipeline components * quick fix for entity linker * fixing types for spancat, textcat, etc * bugfix for tok2vec * type annotations for scorer * add runtime_checkable for Protocol * type and import fixes in tests * mypy fixes for training utilities * few fixes in util * fix import * 🐵 Remove unused `# type: ignore` directives * 🏷 Annotate `Language._components` * 🏷 Annotate `spacy.pipeline.Pipe` * add doc as property to span.pyi * small fixes and cleanup * explicit type annotations instead of via comment Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com> Co-authored-by: svlandeg <sofie.vanlandeghem@gmail.com> Co-authored-by: svlandeg <svlandeg@github.com>
2021-10-14 13:21:40 +00:00
lex.rank = nlp.vocab.vectors.key2row.get(lex.orth, OOV_RANK) # type: ignore[attr-defined]
2020-09-28 13:09:59 +00:00
def init_tok2vec(
2020-09-29 14:47:55 +00:00
nlp: "Language", pretrain_config: Dict[str, Any], init_config: Dict[str, Any]
2020-09-28 13:09:59 +00:00
) -> bool:
# Load pretrained tok2vec weights - cf. CLI command 'pretrain'
P = pretrain_config
2020-09-29 14:47:55 +00:00
I = init_config
2020-09-28 13:09:59 +00:00
weights_data = None
2020-09-29 14:47:55 +00:00
init_tok2vec = ensure_path(I["init_tok2vec"])
2020-09-28 13:09:59 +00:00
if init_tok2vec is not None:
if not init_tok2vec.exists():
err = f"can't find pretrained tok2vec: {init_tok2vec}"
errors = [{"loc": ["initialize", "init_tok2vec"], "msg": err}]
2020-09-28 13:09:59 +00:00
raise ConfigValidationError(config=nlp.config, errors=errors)
with init_tok2vec.open("rb") as file_:
weights_data = file_.read()
if weights_data is not None:
layer = get_tok2vec_ref(nlp, P)
2020-09-28 13:09:59 +00:00
layer.from_bytes(weights_data)
logger.info("Loaded pretrained weights from %s", init_tok2vec)
2020-09-28 13:09:59 +00:00
return True
return False
2020-09-29 08:58:50 +00:00
def convert_vectors(
nlp: "Language",
2020-09-29 08:58:50 +00:00
vectors_loc: Optional[Path],
*,
truncate: int,
prune: int,
name: Optional[str] = None,
Add support for floret vectors (#8909) * 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
2021-10-27 12:08:31 +00:00
mode: str = VectorsMode.default,
attr: str = "ORTH",
2020-09-29 08:58:50 +00:00
) -> None:
vectors_loc = ensure_path(vectors_loc)
if vectors_loc and vectors_loc.parts[-1].endswith(".npz"):
if attr != "ORTH":
raise ValueError(
"ORTH is the only attribute supported for vectors in .npz format."
)
Add support for floret vectors (#8909) * 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
2021-10-27 12:08:31 +00:00
nlp.vocab.vectors = Vectors(
strings=nlp.vocab.strings, data=numpy.load(vectors_loc.open("rb"))
)
2020-09-29 08:58:50 +00:00
for lex in nlp.vocab:
if lex.rank and lex.rank != OOV_RANK:
🏷 Add Mypy check to CI and ignore all existing Mypy errors (#9167) * 🚨 Ignore all existing Mypy errors * 🏗 Add Mypy check to CI * Add types-mock and types-requests as dev requirements * Add additional type ignore directives * Add types packages to dev-only list in reqs test * Add types-dataclasses for python 3.6 * Add ignore to pretrain * 🏷 Improve type annotation on `run_command` helper The `run_command` helper previously declared that it returned an `Optional[subprocess.CompletedProcess]`, but it isn't actually possible for the function to return `None`. These changes modify the type annotation of the `run_command` helper and remove all now-unnecessary `# type: ignore` directives. * 🔧 Allow variable type redefinition in limited contexts These changes modify how Mypy is configured to allow variables to have their type automatically redefined under certain conditions. The Mypy documentation contains the following example: ```python def process(items: List[str]) -> None: # 'items' has type List[str] items = [item.split() for item in items] # 'items' now has type List[List[str]] ... ``` This configuration change is especially helpful in reducing the number of `# type: ignore` directives needed to handle the common pattern of: * Accepting a filepath as a string * Overwriting the variable using `filepath = ensure_path(filepath)` These changes enable redefinition and remove all `# type: ignore` directives rendered redundant by this change. * 🏷 Add type annotation to converters mapping * 🚨 Fix Mypy error in convert CLI argument verification * 🏷 Improve type annotation on `resolve_dot_names` helper * 🏷 Add type annotations for `Vocab` attributes `strings` and `vectors` * 🏷 Add type annotations for more `Vocab` attributes * 🏷 Add loose type annotation for gold data compilation * 🏷 Improve `_format_labels` type annotation * 🏷 Fix `get_lang_class` type annotation * 🏷 Loosen return type of `Language.evaluate` * 🏷 Don't accept `Scorer` in `handle_scores_per_type` * 🏷 Add `string_to_list` overloads * 🏷 Fix non-Optional command-line options * 🙈 Ignore redefinition of `wandb_logger` in `loggers.py` * ➕ Install `typing_extensions` in Python 3.8+ The `typing_extensions` package states that it should be used when "writing code that must be compatible with multiple Python versions". Since SpaCy needs to support multiple Python versions, it should be used when newer `typing` module members are required. One example of this is `Literal`, which is available starting with Python 3.8. Previously SpaCy tried to import `Literal` from `typing`, falling back to `typing_extensions` if the import failed. However, Mypy doesn't seem to be able to understand what `Literal` means when the initial import means. Therefore, these changes modify how `compat` imports `Literal` by always importing it from `typing_extensions`. These changes also modify how `typing_extensions` is installed, so that it is a requirement for all Python versions, including those greater than or equal to 3.8. * 🏷 Improve type annotation for `Language.pipe` These changes add a missing overload variant to the type signature of `Language.pipe`. Additionally, the type signature is enhanced to allow type checkers to differentiate between the two overload variants based on the `as_tuple` parameter. Fixes #8772 * ➖ Don't install `typing-extensions` in Python 3.8+ After more detailed analysis of how to implement Python version-specific type annotations using SpaCy, it has been determined that by branching on a comparison against `sys.version_info` can be statically analyzed by Mypy well enough to enable us to conditionally use `typing_extensions.Literal`. This means that we no longer need to install `typing_extensions` for Python versions greater than or equal to 3.8! 🎉 These changes revert previous changes installing `typing-extensions` regardless of Python version and modify how we import the `Literal` type to ensure that Mypy treats it properly. * resolve mypy errors for Strict pydantic types * refactor code to avoid missing return statement * fix types of convert CLI command * avoid list-set confustion in debug_data * fix typo and formatting * small fixes to avoid type ignores * fix types in profile CLI command and make it more efficient * type fixes in projects CLI * put one ignore back * type fixes for render * fix render types - the sequel * fix BaseDefault in language definitions * fix type of noun_chunks iterator - yields tuple instead of span * fix types in language-specific modules * 🏷 Expand accepted inputs of `get_string_id` `get_string_id` accepts either a string (in which case it returns its ID) or an ID (in which case it immediately returns the ID). These changes extend the type annotation of `get_string_id` to indicate that it can accept either strings or IDs. * 🏷 Handle override types in `combine_score_weights` The `combine_score_weights` function allows users to pass an `overrides` mapping to override data extracted from the `weights` argument. Since it allows `Optional` dictionary values, the return value may also include `Optional` dictionary values. These changes update the type annotations for `combine_score_weights` to reflect this fact. * 🏷 Fix tokenizer serialization method signatures in `DummyTokenizer` * 🏷 Fix redefinition of `wandb_logger` These changes fix the redefinition of `wandb_logger` by giving a separate name to each `WandbLogger` version. For backwards-compatibility, `spacy.train` still exports `wandb_logger_v3` as `wandb_logger` for now. * more fixes for typing in language * type fixes in model definitions * 🏷 Annotate `_RandomWords.probs` as `NDArray` * 🏷 Annotate `tok2vec` layers to help Mypy * 🐛 Fix `_RandomWords.probs` type annotations for Python 3.6 Also remove an import that I forgot to move to the top of the module 😅 * more fixes for matchers and other pipeline components * quick fix for entity linker * fixing types for spancat, textcat, etc * bugfix for tok2vec * type annotations for scorer * add runtime_checkable for Protocol * type and import fixes in tests * mypy fixes for training utilities * few fixes in util * fix import * 🐵 Remove unused `# type: ignore` directives * 🏷 Annotate `Language._components` * 🏷 Annotate `spacy.pipeline.Pipe` * add doc as property to span.pyi * small fixes and cleanup * explicit type annotations instead of via comment Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com> Co-authored-by: svlandeg <sofie.vanlandeghem@gmail.com> Co-authored-by: svlandeg <svlandeg@github.com>
2021-10-14 13:21:40 +00:00
nlp.vocab.vectors.add(lex.orth, row=lex.rank) # type: ignore[attr-defined]
nlp.vocab.deduplicate_vectors()
2020-09-29 08:58:50 +00:00
else:
if vectors_loc:
logger.info("Reading vectors from %s", vectors_loc)
Add support for floret vectors (#8909) * 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
2021-10-27 12:08:31 +00:00
vectors_data, vector_keys, floret_settings = read_vectors(
vectors_loc,
truncate,
mode=mode,
)
logger.info("Loaded vectors from %s", vectors_loc)
2020-09-29 08:58:50 +00:00
else:
vectors_data, vector_keys = (None, None)
Add support for floret vectors (#8909) * 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
2021-10-27 12:08:31 +00:00
if vector_keys is not None and mode != VectorsMode.floret:
2020-09-29 08:58:50 +00:00
for word in vector_keys:
if word not in nlp.vocab:
nlp.vocab[word]
if vectors_data is not None:
Add support for floret vectors (#8909) * 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
2021-10-27 12:08:31 +00:00
if mode == VectorsMode.floret:
nlp.vocab.vectors = Vectors(
strings=nlp.vocab.strings,
data=vectors_data,
attr=attr,
Add support for floret vectors (#8909) * 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
2021-10-27 12:08:31 +00:00
**floret_settings,
)
else:
nlp.vocab.vectors = Vectors(
strings=nlp.vocab.strings,
data=vectors_data,
keys=vector_keys,
attr=attr,
Add support for floret vectors (#8909) * 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
2021-10-27 12:08:31 +00:00
)
nlp.vocab.deduplicate_vectors()
2020-09-29 08:58:50 +00:00
if name is None:
# TODO: Is this correct? Does this matter?
nlp.vocab.vectors.name = f"{nlp.meta['lang']}_{nlp.meta['name']}.vectors"
else:
nlp.vocab.vectors.name = name
nlp.meta["vectors"]["name"] = nlp.vocab.vectors.name
Add support for floret vectors (#8909) * 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
2021-10-27 12:08:31 +00:00
if prune >= 1 and mode != VectorsMode.floret:
2020-09-29 08:58:50 +00:00
nlp.vocab.prune_vectors(prune)
Add support for floret vectors (#8909) * 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
2021-10-27 12:08:31 +00:00
def read_vectors(
vectors_loc: Path, truncate_vectors: int, *, mode: str = VectorsMode.default
):
f = ensure_shape(vectors_loc)
Add support for floret vectors (#8909) * 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
2021-10-27 12:08:31 +00:00
header_parts = next(f).split()
shape = tuple(int(size) for size in header_parts[:2])
floret_settings = {}
if mode == VectorsMode.floret:
if len(header_parts) != 8:
raise ValueError(
"Invalid header for floret vectors. "
"Expected: bucket dim minn maxn hash_count hash_seed BOW EOW"
)
floret_settings = {
"mode": "floret",
"minn": int(header_parts[2]),
"maxn": int(header_parts[3]),
"hash_count": int(header_parts[4]),
"hash_seed": int(header_parts[5]),
"bow": header_parts[6],
"eow": header_parts[7],
}
if truncate_vectors >= 1:
raise ValueError(Errors.E860)
else:
assert len(header_parts) == 2
if truncate_vectors >= 1:
shape = (truncate_vectors, shape[1])
2020-09-29 08:58:50 +00:00
vectors_data = numpy.zeros(shape=shape, dtype="f")
vectors_keys = []
for i, line in enumerate(tqdm.tqdm(f, disable=None)):
2020-09-29 08:58:50 +00:00
line = line.rstrip()
pieces = line.rsplit(" ", vectors_data.shape[1])
word = pieces.pop(0)
if len(pieces) != vectors_data.shape[1]:
raise ValueError(Errors.E094.format(line_num=i, loc=vectors_loc))
vectors_data[i] = numpy.asarray(pieces, dtype="f")
vectors_keys.append(word)
if i == truncate_vectors - 1:
break
Add support for floret vectors (#8909) * 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
2021-10-27 12:08:31 +00:00
return vectors_data, vectors_keys, floret_settings
2020-09-29 08:58:50 +00:00
def open_file(loc: Union[str, Path]) -> IO:
"""Handle .gz, .tar.gz or unzipped files"""
loc = ensure_path(loc)
if tarfile.is_tarfile(str(loc)):
🏷 Add Mypy check to CI and ignore all existing Mypy errors (#9167) * 🚨 Ignore all existing Mypy errors * 🏗 Add Mypy check to CI * Add types-mock and types-requests as dev requirements * Add additional type ignore directives * Add types packages to dev-only list in reqs test * Add types-dataclasses for python 3.6 * Add ignore to pretrain * 🏷 Improve type annotation on `run_command` helper The `run_command` helper previously declared that it returned an `Optional[subprocess.CompletedProcess]`, but it isn't actually possible for the function to return `None`. These changes modify the type annotation of the `run_command` helper and remove all now-unnecessary `# type: ignore` directives. * 🔧 Allow variable type redefinition in limited contexts These changes modify how Mypy is configured to allow variables to have their type automatically redefined under certain conditions. The Mypy documentation contains the following example: ```python def process(items: List[str]) -> None: # 'items' has type List[str] items = [item.split() for item in items] # 'items' now has type List[List[str]] ... ``` This configuration change is especially helpful in reducing the number of `# type: ignore` directives needed to handle the common pattern of: * Accepting a filepath as a string * Overwriting the variable using `filepath = ensure_path(filepath)` These changes enable redefinition and remove all `# type: ignore` directives rendered redundant by this change. * 🏷 Add type annotation to converters mapping * 🚨 Fix Mypy error in convert CLI argument verification * 🏷 Improve type annotation on `resolve_dot_names` helper * 🏷 Add type annotations for `Vocab` attributes `strings` and `vectors` * 🏷 Add type annotations for more `Vocab` attributes * 🏷 Add loose type annotation for gold data compilation * 🏷 Improve `_format_labels` type annotation * 🏷 Fix `get_lang_class` type annotation * 🏷 Loosen return type of `Language.evaluate` * 🏷 Don't accept `Scorer` in `handle_scores_per_type` * 🏷 Add `string_to_list` overloads * 🏷 Fix non-Optional command-line options * 🙈 Ignore redefinition of `wandb_logger` in `loggers.py` * ➕ Install `typing_extensions` in Python 3.8+ The `typing_extensions` package states that it should be used when "writing code that must be compatible with multiple Python versions". Since SpaCy needs to support multiple Python versions, it should be used when newer `typing` module members are required. One example of this is `Literal`, which is available starting with Python 3.8. Previously SpaCy tried to import `Literal` from `typing`, falling back to `typing_extensions` if the import failed. However, Mypy doesn't seem to be able to understand what `Literal` means when the initial import means. Therefore, these changes modify how `compat` imports `Literal` by always importing it from `typing_extensions`. These changes also modify how `typing_extensions` is installed, so that it is a requirement for all Python versions, including those greater than or equal to 3.8. * 🏷 Improve type annotation for `Language.pipe` These changes add a missing overload variant to the type signature of `Language.pipe`. Additionally, the type signature is enhanced to allow type checkers to differentiate between the two overload variants based on the `as_tuple` parameter. Fixes #8772 * ➖ Don't install `typing-extensions` in Python 3.8+ After more detailed analysis of how to implement Python version-specific type annotations using SpaCy, it has been determined that by branching on a comparison against `sys.version_info` can be statically analyzed by Mypy well enough to enable us to conditionally use `typing_extensions.Literal`. This means that we no longer need to install `typing_extensions` for Python versions greater than or equal to 3.8! 🎉 These changes revert previous changes installing `typing-extensions` regardless of Python version and modify how we import the `Literal` type to ensure that Mypy treats it properly. * resolve mypy errors for Strict pydantic types * refactor code to avoid missing return statement * fix types of convert CLI command * avoid list-set confustion in debug_data * fix typo and formatting * small fixes to avoid type ignores * fix types in profile CLI command and make it more efficient * type fixes in projects CLI * put one ignore back * type fixes for render * fix render types - the sequel * fix BaseDefault in language definitions * fix type of noun_chunks iterator - yields tuple instead of span * fix types in language-specific modules * 🏷 Expand accepted inputs of `get_string_id` `get_string_id` accepts either a string (in which case it returns its ID) or an ID (in which case it immediately returns the ID). These changes extend the type annotation of `get_string_id` to indicate that it can accept either strings or IDs. * 🏷 Handle override types in `combine_score_weights` The `combine_score_weights` function allows users to pass an `overrides` mapping to override data extracted from the `weights` argument. Since it allows `Optional` dictionary values, the return value may also include `Optional` dictionary values. These changes update the type annotations for `combine_score_weights` to reflect this fact. * 🏷 Fix tokenizer serialization method signatures in `DummyTokenizer` * 🏷 Fix redefinition of `wandb_logger` These changes fix the redefinition of `wandb_logger` by giving a separate name to each `WandbLogger` version. For backwards-compatibility, `spacy.train` still exports `wandb_logger_v3` as `wandb_logger` for now. * more fixes for typing in language * type fixes in model definitions * 🏷 Annotate `_RandomWords.probs` as `NDArray` * 🏷 Annotate `tok2vec` layers to help Mypy * 🐛 Fix `_RandomWords.probs` type annotations for Python 3.6 Also remove an import that I forgot to move to the top of the module 😅 * more fixes for matchers and other pipeline components * quick fix for entity linker * fixing types for spancat, textcat, etc * bugfix for tok2vec * type annotations for scorer * add runtime_checkable for Protocol * type and import fixes in tests * mypy fixes for training utilities * few fixes in util * fix import * 🐵 Remove unused `# type: ignore` directives * 🏷 Annotate `Language._components` * 🏷 Annotate `spacy.pipeline.Pipe` * add doc as property to span.pyi * small fixes and cleanup * explicit type annotations instead of via comment Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com> Co-authored-by: svlandeg <sofie.vanlandeghem@gmail.com> Co-authored-by: svlandeg <svlandeg@github.com>
2021-10-14 13:21:40 +00:00
return tarfile.open(str(loc), "r:gz") # type: ignore[return-value]
2020-09-29 08:58:50 +00:00
elif loc.parts[-1].endswith("gz"):
🏷 Add Mypy check to CI and ignore all existing Mypy errors (#9167) * 🚨 Ignore all existing Mypy errors * 🏗 Add Mypy check to CI * Add types-mock and types-requests as dev requirements * Add additional type ignore directives * Add types packages to dev-only list in reqs test * Add types-dataclasses for python 3.6 * Add ignore to pretrain * 🏷 Improve type annotation on `run_command` helper The `run_command` helper previously declared that it returned an `Optional[subprocess.CompletedProcess]`, but it isn't actually possible for the function to return `None`. These changes modify the type annotation of the `run_command` helper and remove all now-unnecessary `# type: ignore` directives. * 🔧 Allow variable type redefinition in limited contexts These changes modify how Mypy is configured to allow variables to have their type automatically redefined under certain conditions. The Mypy documentation contains the following example: ```python def process(items: List[str]) -> None: # 'items' has type List[str] items = [item.split() for item in items] # 'items' now has type List[List[str]] ... ``` This configuration change is especially helpful in reducing the number of `# type: ignore` directives needed to handle the common pattern of: * Accepting a filepath as a string * Overwriting the variable using `filepath = ensure_path(filepath)` These changes enable redefinition and remove all `# type: ignore` directives rendered redundant by this change. * 🏷 Add type annotation to converters mapping * 🚨 Fix Mypy error in convert CLI argument verification * 🏷 Improve type annotation on `resolve_dot_names` helper * 🏷 Add type annotations for `Vocab` attributes `strings` and `vectors` * 🏷 Add type annotations for more `Vocab` attributes * 🏷 Add loose type annotation for gold data compilation * 🏷 Improve `_format_labels` type annotation * 🏷 Fix `get_lang_class` type annotation * 🏷 Loosen return type of `Language.evaluate` * 🏷 Don't accept `Scorer` in `handle_scores_per_type` * 🏷 Add `string_to_list` overloads * 🏷 Fix non-Optional command-line options * 🙈 Ignore redefinition of `wandb_logger` in `loggers.py` * ➕ Install `typing_extensions` in Python 3.8+ The `typing_extensions` package states that it should be used when "writing code that must be compatible with multiple Python versions". Since SpaCy needs to support multiple Python versions, it should be used when newer `typing` module members are required. One example of this is `Literal`, which is available starting with Python 3.8. Previously SpaCy tried to import `Literal` from `typing`, falling back to `typing_extensions` if the import failed. However, Mypy doesn't seem to be able to understand what `Literal` means when the initial import means. Therefore, these changes modify how `compat` imports `Literal` by always importing it from `typing_extensions`. These changes also modify how `typing_extensions` is installed, so that it is a requirement for all Python versions, including those greater than or equal to 3.8. * 🏷 Improve type annotation for `Language.pipe` These changes add a missing overload variant to the type signature of `Language.pipe`. Additionally, the type signature is enhanced to allow type checkers to differentiate between the two overload variants based on the `as_tuple` parameter. Fixes #8772 * ➖ Don't install `typing-extensions` in Python 3.8+ After more detailed analysis of how to implement Python version-specific type annotations using SpaCy, it has been determined that by branching on a comparison against `sys.version_info` can be statically analyzed by Mypy well enough to enable us to conditionally use `typing_extensions.Literal`. This means that we no longer need to install `typing_extensions` for Python versions greater than or equal to 3.8! 🎉 These changes revert previous changes installing `typing-extensions` regardless of Python version and modify how we import the `Literal` type to ensure that Mypy treats it properly. * resolve mypy errors for Strict pydantic types * refactor code to avoid missing return statement * fix types of convert CLI command * avoid list-set confustion in debug_data * fix typo and formatting * small fixes to avoid type ignores * fix types in profile CLI command and make it more efficient * type fixes in projects CLI * put one ignore back * type fixes for render * fix render types - the sequel * fix BaseDefault in language definitions * fix type of noun_chunks iterator - yields tuple instead of span * fix types in language-specific modules * 🏷 Expand accepted inputs of `get_string_id` `get_string_id` accepts either a string (in which case it returns its ID) or an ID (in which case it immediately returns the ID). These changes extend the type annotation of `get_string_id` to indicate that it can accept either strings or IDs. * 🏷 Handle override types in `combine_score_weights` The `combine_score_weights` function allows users to pass an `overrides` mapping to override data extracted from the `weights` argument. Since it allows `Optional` dictionary values, the return value may also include `Optional` dictionary values. These changes update the type annotations for `combine_score_weights` to reflect this fact. * 🏷 Fix tokenizer serialization method signatures in `DummyTokenizer` * 🏷 Fix redefinition of `wandb_logger` These changes fix the redefinition of `wandb_logger` by giving a separate name to each `WandbLogger` version. For backwards-compatibility, `spacy.train` still exports `wandb_logger_v3` as `wandb_logger` for now. * more fixes for typing in language * type fixes in model definitions * 🏷 Annotate `_RandomWords.probs` as `NDArray` * 🏷 Annotate `tok2vec` layers to help Mypy * 🐛 Fix `_RandomWords.probs` type annotations for Python 3.6 Also remove an import that I forgot to move to the top of the module 😅 * more fixes for matchers and other pipeline components * quick fix for entity linker * fixing types for spancat, textcat, etc * bugfix for tok2vec * type annotations for scorer * add runtime_checkable for Protocol * type and import fixes in tests * mypy fixes for training utilities * few fixes in util * fix import * 🐵 Remove unused `# type: ignore` directives * 🏷 Annotate `Language._components` * 🏷 Annotate `spacy.pipeline.Pipe` * add doc as property to span.pyi * small fixes and cleanup * explicit type annotations instead of via comment Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com> Co-authored-by: svlandeg <sofie.vanlandeghem@gmail.com> Co-authored-by: svlandeg <svlandeg@github.com>
2021-10-14 13:21:40 +00:00
return (line.decode("utf8") for line in gzip.open(str(loc), "r")) # type: ignore[return-value]
2020-09-29 08:58:50 +00:00
elif loc.parts[-1].endswith("zip"):
zip_file = zipfile.ZipFile(str(loc))
names = zip_file.namelist()
file_ = zip_file.open(names[0])
🏷 Add Mypy check to CI and ignore all existing Mypy errors (#9167) * 🚨 Ignore all existing Mypy errors * 🏗 Add Mypy check to CI * Add types-mock and types-requests as dev requirements * Add additional type ignore directives * Add types packages to dev-only list in reqs test * Add types-dataclasses for python 3.6 * Add ignore to pretrain * 🏷 Improve type annotation on `run_command` helper The `run_command` helper previously declared that it returned an `Optional[subprocess.CompletedProcess]`, but it isn't actually possible for the function to return `None`. These changes modify the type annotation of the `run_command` helper and remove all now-unnecessary `# type: ignore` directives. * 🔧 Allow variable type redefinition in limited contexts These changes modify how Mypy is configured to allow variables to have their type automatically redefined under certain conditions. The Mypy documentation contains the following example: ```python def process(items: List[str]) -> None: # 'items' has type List[str] items = [item.split() for item in items] # 'items' now has type List[List[str]] ... ``` This configuration change is especially helpful in reducing the number of `# type: ignore` directives needed to handle the common pattern of: * Accepting a filepath as a string * Overwriting the variable using `filepath = ensure_path(filepath)` These changes enable redefinition and remove all `# type: ignore` directives rendered redundant by this change. * 🏷 Add type annotation to converters mapping * 🚨 Fix Mypy error in convert CLI argument verification * 🏷 Improve type annotation on `resolve_dot_names` helper * 🏷 Add type annotations for `Vocab` attributes `strings` and `vectors` * 🏷 Add type annotations for more `Vocab` attributes * 🏷 Add loose type annotation for gold data compilation * 🏷 Improve `_format_labels` type annotation * 🏷 Fix `get_lang_class` type annotation * 🏷 Loosen return type of `Language.evaluate` * 🏷 Don't accept `Scorer` in `handle_scores_per_type` * 🏷 Add `string_to_list` overloads * 🏷 Fix non-Optional command-line options * 🙈 Ignore redefinition of `wandb_logger` in `loggers.py` * ➕ Install `typing_extensions` in Python 3.8+ The `typing_extensions` package states that it should be used when "writing code that must be compatible with multiple Python versions". Since SpaCy needs to support multiple Python versions, it should be used when newer `typing` module members are required. One example of this is `Literal`, which is available starting with Python 3.8. Previously SpaCy tried to import `Literal` from `typing`, falling back to `typing_extensions` if the import failed. However, Mypy doesn't seem to be able to understand what `Literal` means when the initial import means. Therefore, these changes modify how `compat` imports `Literal` by always importing it from `typing_extensions`. These changes also modify how `typing_extensions` is installed, so that it is a requirement for all Python versions, including those greater than or equal to 3.8. * 🏷 Improve type annotation for `Language.pipe` These changes add a missing overload variant to the type signature of `Language.pipe`. Additionally, the type signature is enhanced to allow type checkers to differentiate between the two overload variants based on the `as_tuple` parameter. Fixes #8772 * ➖ Don't install `typing-extensions` in Python 3.8+ After more detailed analysis of how to implement Python version-specific type annotations using SpaCy, it has been determined that by branching on a comparison against `sys.version_info` can be statically analyzed by Mypy well enough to enable us to conditionally use `typing_extensions.Literal`. This means that we no longer need to install `typing_extensions` for Python versions greater than or equal to 3.8! 🎉 These changes revert previous changes installing `typing-extensions` regardless of Python version and modify how we import the `Literal` type to ensure that Mypy treats it properly. * resolve mypy errors for Strict pydantic types * refactor code to avoid missing return statement * fix types of convert CLI command * avoid list-set confustion in debug_data * fix typo and formatting * small fixes to avoid type ignores * fix types in profile CLI command and make it more efficient * type fixes in projects CLI * put one ignore back * type fixes for render * fix render types - the sequel * fix BaseDefault in language definitions * fix type of noun_chunks iterator - yields tuple instead of span * fix types in language-specific modules * 🏷 Expand accepted inputs of `get_string_id` `get_string_id` accepts either a string (in which case it returns its ID) or an ID (in which case it immediately returns the ID). These changes extend the type annotation of `get_string_id` to indicate that it can accept either strings or IDs. * 🏷 Handle override types in `combine_score_weights` The `combine_score_weights` function allows users to pass an `overrides` mapping to override data extracted from the `weights` argument. Since it allows `Optional` dictionary values, the return value may also include `Optional` dictionary values. These changes update the type annotations for `combine_score_weights` to reflect this fact. * 🏷 Fix tokenizer serialization method signatures in `DummyTokenizer` * 🏷 Fix redefinition of `wandb_logger` These changes fix the redefinition of `wandb_logger` by giving a separate name to each `WandbLogger` version. For backwards-compatibility, `spacy.train` still exports `wandb_logger_v3` as `wandb_logger` for now. * more fixes for typing in language * type fixes in model definitions * 🏷 Annotate `_RandomWords.probs` as `NDArray` * 🏷 Annotate `tok2vec` layers to help Mypy * 🐛 Fix `_RandomWords.probs` type annotations for Python 3.6 Also remove an import that I forgot to move to the top of the module 😅 * more fixes for matchers and other pipeline components * quick fix for entity linker * fixing types for spancat, textcat, etc * bugfix for tok2vec * type annotations for scorer * add runtime_checkable for Protocol * type and import fixes in tests * mypy fixes for training utilities * few fixes in util * fix import * 🐵 Remove unused `# type: ignore` directives * 🏷 Annotate `Language._components` * 🏷 Annotate `spacy.pipeline.Pipe` * add doc as property to span.pyi * small fixes and cleanup * explicit type annotations instead of via comment Co-authored-by: Adriane Boyd <adrianeboyd@gmail.com> Co-authored-by: svlandeg <sofie.vanlandeghem@gmail.com> Co-authored-by: svlandeg <svlandeg@github.com>
2021-10-14 13:21:40 +00:00
return (line.decode("utf8") for line in file_) # type: ignore[return-value]
2020-09-29 08:58:50 +00:00
else:
return loc.open("r", encoding="utf8")
def ensure_shape(vectors_loc):
2020-09-29 08:58:50 +00:00
"""Ensure that the first line of the data is the vectors shape.
If it's not, we read in the data and output the shape as the first result,
so that the reader doesn't have to deal with the problem.
"""
lines = open_file(vectors_loc)
2020-09-29 08:58:50 +00:00
first_line = next(lines)
try:
Add support for floret vectors (#8909) * 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
2021-10-27 12:08:31 +00:00
shape = tuple(int(size) for size in first_line.split()[:2])
2020-09-29 08:58:50 +00:00
except ValueError:
shape = None
if shape is not None:
# All good, give the data
yield first_line
yield from lines
else:
# Figure out the shape, make it the first value, and then give the
# rest of the data.
width = len(first_line.split()) - 1
length = 1
for _ in lines:
length += 1
2020-09-29 08:58:50 +00:00
yield f"{length} {width}"
# Reading the lines in again from file. This to avoid having to
# store all the results in a list in memory
lines2 = open_file(vectors_loc)
yield from lines2
lines2.close()
lines.close()