explosion/spaCy
1
1
Fork 0
mirror of https://github.com/explosion/spaCy.git synced 2025-01-13 18:56:36 +03:00
Code Issues Packages Projects Releases Wiki Activity

Remove redundant PrecomputableMaxouts class

Browse Source
This commit is contained in:
Matthew Honnibal 2017-10-19 20:27:34 +02:00
parent a17a1b60c7
commit a8850b4282
1 changed files with 15 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

32
spacy/syntax/nn_parser.pyx
View File

@ -47,7 +47,7 @@ from thinc.neural.util import get_array_module
from .. import util from .. import util
from ..util import get_async, get_cuda_stream from ..util import get_async, get_cuda_stream
from .._ml import zero_init, PrecomputableAffine, PrecomputableMaxouts from .._ml import zero_init, PrecomputableAffine
from .._ml import Tok2Vec, doc2feats, rebatch, fine_tune from .._ml import Tok2Vec, doc2feats, rebatch, fine_tune
from .._ml import Residual, drop_layer, flatten from .._ml import Residual, drop_layer, flatten
from .._ml import link_vectors_to_models from .._ml import link_vectors_to_models
@ -153,8 +153,7 @@ cdef class precompute_hiddens:
state_vector, bp_nonlinearity = self._nonlinearity(state_vector) state_vector, bp_nonlinearity = self._nonlinearity(state_vector)
def backward(d_state_vector, sgd=None): def backward(d_state_vector, sgd=None):
if bp_nonlinearity is not None: d_state_vector = bp_nonlinearity(d_state_vector, sgd)
d_state_vector = bp_nonlinearity(d_state_vector, sgd)
# This will usually be on GPU # This will usually be on GPU
if not isinstance(d_state_vector, self.ops.xp.ndarray): if not isinstance(d_state_vector, self.ops.xp.ndarray):
d_state_vector = self.ops.xp.array(d_state_vector) d_state_vector = self.ops.xp.array(d_state_vector)
@ -165,14 +164,18 @@ cdef class precompute_hiddens:
def _nonlinearity(self, state_vector): def _nonlinearity(self, state_vector):
if self.nP == 1: if self.nP == 1:
mask = state_vector >= 0. mask = state_vector >= 0.
return state_vector * mask, lambda dY, sgd=None: dY * mask state_vector *= mask
state_vector = state_vector.reshape( else:
(state_vector.shape[0], state_vector.shape[1]//self.nP, self.nP)) state_vector = state_vector.reshape(
best, which = self.ops.maxout(state_vector) (state_vector.shape[0], self.nO, self.nP))
state_vector, mask = self.ops.maxout(state_vector)
def backprop_maxout(d_best, sgd=None): def backprop_nonlinearity(d_best, sgd=None):
return self.ops.backprop_maxout(d_best, which, self.nP) if self.nP == 1:
return best, backprop_maxout return d_best * mask
else:
return self.ops.backprop_maxout(d_best, mask, self.nP)
return state_vector, backprop_nonlinearity
cdef void sum_state_features(float* output, cdef void sum_state_features(float* output,
@ -262,13 +265,8 @@ cdef class Parser:
tok2vec = Tok2Vec(token_vector_width, embed_size, tok2vec = Tok2Vec(token_vector_width, embed_size,
pretrained_dims=cfg.get('pretrained_dims', 0)) pretrained_dims=cfg.get('pretrained_dims', 0))
tok2vec = chain(tok2vec, flatten) tok2vec = chain(tok2vec, flatten)
if parser_maxout_pieces >= 2: lower = PrecomputableAffine(hidden_width * parser_maxout_pieces,
lower = PrecomputableMaxouts(hidden_width if depth >= 1 else nr_class, nF=cls.nr_feature, nI=token_vector_width)
nF=cls.nr_feature, nP=parser_maxout_pieces,
nI=token_vector_width)
else:
lower = PrecomputableAffine(hidden_width if depth >= 1 else nr_class,
nF=cls.nr_feature, nI=token_vector_width)
with Model.use_device('cpu'): with Model.use_device('cpu'):
upper = chain( upper = chain(