mirror of
https://github.com/explosion/spaCy.git
synced 2024-11-11 20:28:20 +03:00
333b1a308b
* Draft layer for BILUO actions * Fixes to biluo layer * WIP on BILUO layer * Add tests for BILUO layer * Format * Fix transitions * Update test * Link in the simple_ner * Update BILUO tagger * Update __init__ * Import simple_ner * Update test * Import * Add files * Add config * Fix label passing for BILUO and tagger * Fix label handling for simple_ner component * Update simple NER test * Update config * Hack train script * Update BILUO layer * Fix SimpleNER component * Update train_from_config * Add biluo_to_iob helper * Add IOB layer * Add IOBTagger model * Update biluo layer * Update SimpleNER tagger * Update BILUO * Read random seed in train-from-config * Update use of normal_init * Fix normalization of gradient in SimpleNER * Update IOBTagger * Remove print * Tweak masking in BILUO * Add dropout in SimpleNER * Update thinc * Tidy up simple_ner * Fix biluo model * Unhack train-from-config * Update setup.cfg and requirements * Add tb_framework.py for parser model * Try to avoid memory leak in BILUO * Move ParserModel into spacy.ml, avoid need for subclass. * Use updated parser model * Remove incorrect call to model.initializre in PrecomputableAffine * Update parser model * Avoid divide by zero in tagger * Add extra dropout layer in tagger * Refine minibatch_by_words function to avoid oom * Fix parser model after refactor * Try to avoid div-by-zero in SimpleNER * Fix infinite loop in minibatch_by_words * Use SequenceCategoricalCrossentropy in Tagger * Fix parser model when hidden layer * Remove extra dropout from tagger * Add extra nan check in tagger * Fix thinc version * Update tests and imports * Fix test * Update test * Update tests * Fix tests * Fix test Co-authored-by: Ines Montani <ines@ines.io>
87 lines
2.8 KiB
Python
87 lines
2.8 KiB
Python
from thinc.api import Model, noop, use_ops, Linear
|
|
from ..syntax._parser_model import ParserStepModel
|
|
|
|
|
|
def TransitionModel(tok2vec, lower, upper, unseen_classes=set()):
|
|
"""Set up a stepwise transition-based model"""
|
|
if upper is None:
|
|
has_upper = False
|
|
upper = noop()
|
|
else:
|
|
has_upper = True
|
|
# don't define nO for this object, because we can't dynamically change it
|
|
return Model(
|
|
name="parser_model",
|
|
forward=forward,
|
|
dims={"nI": tok2vec.get_dim("nI") if tok2vec.has_dim("nI") else None},
|
|
layers=[tok2vec, lower, upper],
|
|
refs={"tok2vec": tok2vec, "lower": lower, "upper": upper},
|
|
init=init,
|
|
attrs={
|
|
"has_upper": has_upper,
|
|
"unseen_classes": set(unseen_classes),
|
|
"resize_output": resize_output
|
|
}
|
|
)
|
|
|
|
|
|
def forward(model, X, is_train):
|
|
step_model = ParserStepModel(
|
|
X,
|
|
model.layers,
|
|
unseen_classes=model.attrs["unseen_classes"],
|
|
train=is_train,
|
|
has_upper=model.attrs["has_upper"]
|
|
)
|
|
|
|
return step_model, step_model.finish_steps
|
|
|
|
|
|
def init(model, X=None, Y=None):
|
|
tok2vec = model.get_ref("tok2vec").initialize()
|
|
lower = model.get_ref("lower").initialize(X=X)
|
|
if model.attrs["has_upper"]:
|
|
statevecs = model.ops.alloc2f(2, lower.get_dim("nO"))
|
|
model.get_ref("upper").initialize(X=statevecs)
|
|
|
|
|
|
def resize_output(model, new_nO):
|
|
tok2vec = model.get_ref("tok2vec")
|
|
lower = model.get_ref("lower")
|
|
upper = model.get_ref("upper")
|
|
if not model.attrs["has_upper"]:
|
|
if lower.has_dim("nO") is None:
|
|
lower.set_dim("nO", new_nO)
|
|
return
|
|
elif upper.has_dim("nO") is None:
|
|
upper.set_dim("nO", new_nO)
|
|
return
|
|
elif new_nO == upper.get_dim("nO"):
|
|
return
|
|
smaller = upper
|
|
nI = None
|
|
if smaller.has_dim("nI"):
|
|
nI = smaller.get_dim("nI")
|
|
with use_ops('numpy'):
|
|
larger = Linear(nO=new_nO, nI=nI)
|
|
larger.init = smaller.init
|
|
# it could be that the model is not initialized yet, then skip this bit
|
|
if nI:
|
|
larger_W = larger.ops.alloc2f(new_nO, nI)
|
|
larger_b = larger.ops.alloc1f(new_nO)
|
|
smaller_W = smaller.get_param("W")
|
|
smaller_b = smaller.get_param("b")
|
|
# Weights are stored in (nr_out, nr_in) format, so we're basically
|
|
# just adding rows here.
|
|
if smaller.has_dim("nO"):
|
|
larger_W[:smaller.get_dim("nO")] = smaller_W
|
|
larger_b[:smaller.get_dim("nO")] = smaller_b
|
|
for i in range(smaller.get_dim("nO"), new_nO):
|
|
model.attrs["unseen_classes"].add(i)
|
|
|
|
larger.set_param("W", larger_W)
|
|
larger.set_param("b", larger_b)
|
|
model._layers[-1] = larger
|
|
model.set_ref("upper", larger)
|
|
return model
|