Rooba
/
spaCy
mirror of https://github.com/explosion/spaCy.git
1
1
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Use PretrainableMaxouts

This commit is contained in:
Matthew Honnibal 2017-05-08 14:24:55 +02:00
parent 807cb2e370
commit 8d2eab74da
1 changed files with 3 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
spacy/syntax/parser.pyx
View File

@ -32,7 +32,7 @@ from preshed.maps cimport map_get
from thinc.api import layerize, chain from thinc.api import layerize, chain
from thinc.neural import Model, Maxout from thinc.neural import Model, Maxout
from .._ml import PrecomputableAffine from .._ml import PrecomputableAffine, PrecomputableMaxouts
from . import _parse_features from . import _parse_features
from ._parse_features cimport CONTEXT_SIZE from ._parse_features cimport CONTEXT_SIZE
from ._parse_features cimport fill_context from ._parse_features cimport fill_context
@ -93,7 +93,7 @@ def get_greedy_model_for_batch(tokvecs, TransitionSystem moves, upper_model, low
for i in range(len(states)): for i in range(len(states)):
for j, tok_i in enumerate(adjusted_ids[i]): for j, tok_i in enumerate(adjusted_ids[i]):
if tok_i >= 0: if tok_i >= 0:
features[i] += cached[tok_i, j] features[i] += cached[j, tok_i]
scores, bp_scores = upper_model.begin_update(features, drop=drop) scores, bp_scores = upper_model.begin_update(features, drop=drop)
scores = upper_model.ops.relu(scores) scores = upper_model.ops.relu(scores)
@ -222,7 +222,7 @@ cdef class Parser:
nr_context_tokens = StateClass.nr_context_tokens(nF, nB, nS, nL, nR) nr_context_tokens = StateClass.nr_context_tokens(nF, nB, nS, nL, nR)
upper = chain(Maxout(width, width), Maxout(self.moves.n_moves, width)) upper = chain(Maxout(width, width), Maxout(self.moves.n_moves, width))
lower = PrecomputableAffine(width, nF=nr_context_tokens, nI=width) lower = PrecomputableMaxouts(width, nF=nr_context_tokens, nI=width)
return upper, lower return upper, lower
def __call__(self, Doc tokens): def __call__(self, Doc tokens):