explosion/spaCy
1
1
Fork 0
mirror of https://github.com/explosion/spaCy.git synced 2025-07-15 18:52:29 +03:00
Code Issues Packages Projects Releases Wiki Activity

add label smoothing

Browse Source
This commit is contained in:
vinit 2023-02-17 00:22:25 +05:30
parent 80bc140533
commit 8be0e39a75
2 changed files with 21 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
spacy/pipeline/tagger.pyx
View File

@ -45,7 +45,7 @@ DEFAULT_TAGGER_MODEL = Config().from_str(default_model_config)["model"]
@Language.factory( @Language.factory(
"tagger", "tagger",
assigns=["token.tag"], assigns=["token.tag"],
default_config={"model": DEFAULT_TAGGER_MODEL, "overwrite": False, "scorer": {"@scorers": "spacy.tagger_scorer.v1"}, "neg_prefix": "!"}, default_config={"model": DEFAULT_TAGGER_MODEL, "overwrite": False, "scorer": {"@scorers": "spacy.tagger_scorer.v1"}, "neg_prefix": "!", "label_smoothing": 0.0},
default_score_weights={"tag_acc": 1.0}, default_score_weights={"tag_acc": 1.0},
) )
def make_tagger( def make_tagger(
@ -55,6 +55,7 @@ def make_tagger(
overwrite: bool, overwrite: bool,
scorer: Optional[Callable], scorer: Optional[Callable],
neg_prefix: str, neg_prefix: str,
label_smoothing: float,
): ):
"""Construct a part-of-speech tagger component. """Construct a part-of-speech tagger component.
@ -63,7 +64,7 @@ def make_tagger(
in size, and be normalized as probabilities (all scores between 0 and 1, in size, and be normalized as probabilities (all scores between 0 and 1,
with the rows summing to 1). with the rows summing to 1).
""" """
return Tagger(nlp.vocab, model, name, overwrite=overwrite, scorer=scorer, neg_prefix=neg_prefix) return Tagger(nlp.vocab, model, name, overwrite=overwrite, scorer=scorer, neg_prefix=neg_prefix, label_smoothing=label_smoothing)
def tagger_score(examples, **kwargs): def tagger_score(examples, **kwargs):
@ -89,6 +90,7 @@ class Tagger(TrainablePipe):
overwrite=BACKWARD_OVERWRITE, overwrite=BACKWARD_OVERWRITE,
scorer=tagger_score, scorer=tagger_score,
neg_prefix="!", neg_prefix="!",
label_smoothing=0.0,
): ):
"""Initialize a part-of-speech tagger. """Initialize a part-of-speech tagger.
@ -105,7 +107,7 @@ class Tagger(TrainablePipe):
self.model = model self.model = model
self.name = name self.name = name
self._rehearsal_model = None self._rehearsal_model = None
cfg = {"labels": [], "overwrite": overwrite, "neg_prefix": neg_prefix} cfg = {"labels": [], "overwrite": overwrite, "neg_prefix": neg_prefix, "label_smoothing": label_smoothing}
self.cfg = dict(sorted(cfg.items())) self.cfg = dict(sorted(cfg.items()))
self.scorer = scorer self.scorer = scorer
@ -256,7 +258,8 @@ class Tagger(TrainablePipe):
DOCS: https://spacy.io/api/tagger#get_loss DOCS: https://spacy.io/api/tagger#get_loss
""" """
validate_examples(examples, "Tagger.get_loss") validate_examples(examples, "Tagger.get_loss")
loss_func = SequenceCategoricalCrossentropy(names=self.labels, normalize=False, neg_prefix=self.cfg["neg_prefix"]) # label_smoothing = 0.1 if self.cfg["label_smoothing"] else 0.0
loss_func = SequenceCategoricalCrossentropy(names=self.labels, normalize=False, neg_prefix=self.cfg["neg_prefix"], label_smoothing=self.cfg["label_smoothing"])
# Convert empty tag "" to missing value None so that both misaligned # Convert empty tag "" to missing value None so that both misaligned
# tokens and tokens with missing annotation have the default missing # tokens and tokens with missing annotation have the default missing
# value None. # value None.

14
spacy/tests/pipeline/test_tagger.py
View File

@ -67,6 +67,20 @@ PARTIAL_DATA = [
] ]
def test_label_smoothing():
nlp = Language()
tagger = nlp.add_pipe("tagger", config=dict(label_smoothing=True))
train_examples = []
for tag in TAGS:
tagger.add_label(tag)
for t in TRAIN_DATA:
train_examples.append(Example.from_dict(nlp.make_doc(t[0]), t[1]))
optimizer = nlp.initialize(get_examples=lambda: train_examples)
for i in range(1):
losses = {}
nlp.update(train_examples, sgd=optimizer, losses=losses)
def test_no_label(): def test_no_label():
nlp = Language() nlp = Language()
nlp.add_pipe("tagger") nlp.add_pipe("tagger")