explosion/spaCy
1
1
Fork 0
mirror of https://github.com/explosion/spaCy.git synced 2025-05-03 15:23:41 +03:00
Code Issues Packages Projects Releases Wiki Activity

Improve Tensorizer

Browse Source
This commit is contained in:
Matthew Honnibal 2018-11-03 10:52:50 +00:00
parent baf7feae68
commit dac3f1b280
1 changed files with 5 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
spacy/pipeline.pyx
View File

@ -31,7 +31,7 @@ from .matcher import Matcher
from .matcher import Matcher, PhraseMatcher from .matcher import Matcher, PhraseMatcher
from .tokens.span import Span from .tokens.span import Span
from .attrs import POS from .attrs import POS, ID
from .parts_of_speech import X from .parts_of_speech import X
from ._ml import Tok2Vec, build_text_classifier, build_tagger_model from ._ml import Tok2Vec, build_text_classifier, build_tagger_model
from ._ml import link_vectors_to_models, zero_init, flatten from ._ml import link_vectors_to_models, zero_init, flatten
@ -434,7 +434,7 @@ class Tensorizer(Pipe):
name = 'tensorizer' name = 'tensorizer'
@classmethod @classmethod
def Model(cls, output_size=300, input_size=128, **cfg): def Model(cls, output_size=300, **cfg):
"""Create a new statistical model for the class. """Create a new statistical model for the class.
width (int): Output size of the model. width (int): Output size of the model.
@ -442,6 +442,7 @@ class Tensorizer(Pipe):
**cfg: Config parameters. **cfg: Config parameters.
RETURNS (Model): A `thinc.neural.Model` or similar instance. RETURNS (Model): A `thinc.neural.Model` or similar instance.
""" """
input_size = util.env_opt('token_vector_width', cfg.get('input_size', 128))
return zero_init(Affine(output_size, input_size)) return zero_init(Affine(output_size, input_size))
def __init__(self, vocab, model=True, **cfg): def __init__(self, vocab, model=True, **cfg):
@ -540,12 +541,8 @@ class Tensorizer(Pipe):
return loss return loss
def get_loss(self, docs, golds, prediction): def get_loss(self, docs, golds, prediction):
target = [] ids = self.model.ops.flatten([doc.to_array(ID).ravel() for doc in docs])
i = 0 target = self.vocab.vectors.data[ids]
for doc in docs:
vectors = self.model.ops.xp.vstack([w.vector for w in doc])
target.append(vectors)
target = self.model.ops.xp.vstack(target)
d_scores = (prediction - target) d_scores = (prediction - target)
loss = (d_scores**2).sum() loss = (d_scores**2).sum()
return loss, d_scores return loss, d_scores