spaCy/spacy/tests/pipeline/test_annotates_on_update.py
Daniël de Kok eec5ccd72f
Language.update: ensure that tok2vec gets updated (#12136)
* `Language.update`: ensure that tok2vec gets updated

The components in a pipeline can be updated independently. However,
tok2vec implementations are an exception to this, since they depend on
listeners for their gradients. The update method of a tok2vec
implementation computes the tok2vec forward and passes this along with a
backprop function to the listeners. This backprop function accumulates
gradients for all the listeners. There are two ways in which the
accumulated gradients can be used to update the tok2vec weights:

1. Call the `finish_update` method of tok2vec *after* the `update`
   method is called on all of the pipes that use a tok2vec listener.
2. Pass an optimizer to the `update` method of tok2vec. In this
   case, tok2vec will give the last listener a special backprop
   function that calls `finish_update` on the tok2vec.

Unfortunately, `Language.update` did neither of these. Instead, it
immediately called `finish_update` on every pipe after `update`. As a
result, the tok2vec weights are updated when no gradients have been
accumulated from listeners yet. And the gradients of the listeners are
only used in the next call to `Language.update` (when `finish_update` is
called on tok2vec again).

This change fixes this issue by passing the optimizer to the `update`
method of trainable pipes, leading to use of the second strategy
outlined above.

The main updating loop in `Language.update` is also simplified by using
the `TrainableComponent` protocol consistently.

* Train loop: `sgd` is `Optional[Optimizer]`, do not pass false

* Language.update: call pipe finish_update after all pipe updates

This does correct and fast updates if multiple components update the
same parameters.

* Add comment why we moved `finish_update` to a separate loop
2023-02-03 15:22:25 +01:00

121 lines
3.3 KiB
Python

from typing import Callable, Iterable, Iterator
import pytest
from thinc.api import Config
from spacy.language import Language
from spacy.training import Example
from spacy.training.loop import train
from spacy.lang.en import English
from spacy.util import registry, load_model_from_config
@pytest.fixture
def config_str():
return """
[nlp]
lang = "en"
pipeline = ["sentencizer","assert_sents"]
disabled = []
before_creation = null
after_creation = null
after_pipeline_creation = null
batch_size = 1000
tokenizer = {"@tokenizers":"spacy.Tokenizer.v1"}
[components]
[components.assert_sents]
factory = "assert_sents"
[components.sentencizer]
factory = "sentencizer"
punct_chars = null
[training]
dev_corpus = "corpora.dev"
train_corpus = "corpora.train"
annotating_components = ["sentencizer"]
max_steps = 2
[corpora]
[corpora.dev]
@readers = "unannotated_corpus"
[corpora.train]
@readers = "unannotated_corpus"
"""
def test_annotates_on_update():
# The custom component checks for sentence annotation
@Language.factory("assert_sents", default_config={})
def assert_sents(nlp, name):
return AssertSents(name)
class AssertSents:
model = None
is_trainable = True
def __init__(self, name, **cfg):
self.name = name
def __call__(self, doc):
if not doc.has_annotation("SENT_START"):
raise ValueError("No sents")
return doc
def update(self, examples, *, drop=0.0, sgd=None, losses=None):
losses.setdefault(self.name, 0.0)
for example in examples:
if not example.predicted.has_annotation("SENT_START"):
raise ValueError("No sents")
return losses
def finish_update(self, sgd=None):
pass
nlp = English()
nlp.add_pipe("sentencizer")
nlp.add_pipe("assert_sents")
# When the pipeline runs, annotations are set
nlp("This is a sentence.")
examples = []
for text in ["a a", "b b", "c c"]:
examples.append(Example(nlp.make_doc(text), nlp(text)))
for example in examples:
assert not example.predicted.has_annotation("SENT_START")
# If updating without setting annotations, assert_sents will raise an error
with pytest.raises(ValueError):
nlp.update(examples)
# Updating while setting annotations for the sentencizer succeeds
nlp.update(examples, annotates=["sentencizer"])
def test_annotating_components_from_config(config_str):
@registry.readers("unannotated_corpus")
def create_unannotated_corpus() -> Callable[[Language], Iterable[Example]]:
return UnannotatedCorpus()
class UnannotatedCorpus:
def __call__(self, nlp: Language) -> Iterator[Example]:
for text in ["a a", "b b", "c c"]:
doc = nlp.make_doc(text)
yield Example(doc, doc)
orig_config = Config().from_str(config_str)
nlp = load_model_from_config(orig_config, auto_fill=True, validate=True)
assert nlp.config["training"]["annotating_components"] == ["sentencizer"]
train(nlp)
nlp.config["training"]["annotating_components"] = []
with pytest.raises(ValueError):
train(nlp)