explosion/spaCy
1
1
Fork 0
mirror of https://github.com/explosion/spaCy.git synced 2025-01-24 00:04:15 +03:00
Code Issues Packages Projects Releases Wiki Activity

Update tok2vec layer

Browse Source
This commit is contained in:
Matthew Honnibal 2020-07-29 13:38:41 +02:00
parent 5ae8628571
commit 07b47eaac8
1 changed files with 20 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
spacy/ml/models/tok2vec.py
View File

@ -107,11 +107,9 @@ def MultiHashEmbed(
) )
else: else:
model = chain( model = chain(
chain(
FeatureExtractor(cols), FeatureExtractor(cols),
list2ragged(), list2ragged(),
with_array(concatenate(*embeddings)), with_array(concatenate(*embeddings)),
),
with_array(Maxout(width, concat_size, nP=3, dropout=0.0, normalize=True)), with_array(Maxout(width, concat_size, nP=3, dropout=0.0, normalize=True)),
ragged2list(), ragged2list(),
) )
@ -120,14 +118,18 @@ def MultiHashEmbed(
@registry.architectures.register("spacy.CharacterEmbed.v1") @registry.architectures.register("spacy.CharacterEmbed.v1")
def CharacterEmbed(width: int, rows: int, nM: int, nC: int): def CharacterEmbed(width: int, rows: int, nM: int, nC: int):
model = concatenate( model = chain(
_character_embed.CharacterEmbed(nM=nM, nC=nC), concatenate(
chain(_character_embed.CharacterEmbed(nM=nM, nC=nC), list2ragged()),
chain( chain(
FeatureExtractor([NORM]), FeatureExtractor([NORM]),
list2ragged(),
with_array(HashEmbed(nO=width, nV=rows, column=0, seed=5)) with_array(HashEmbed(nO=width, nV=rows, column=0, seed=5))
) )
),
with_array(Maxout(width, nM * nC + width, nP=3, normalize=True, dropout=0.0)),
ragged2list()
) )
model.set_dim("nO", nM * nC + width)
return model return model
@ -153,8 +155,12 @@ def MaxoutWindowEncoder(width: int, window_size: int, maxout_pieces: int, depth:
def MishWindowEncoder(width, window_size, depth): def MishWindowEncoder(width, window_size, depth):
cnn = chain( cnn = chain(
expand_window(window_size=window_size), expand_window(window_size=window_size),
Mish(nO=width, nI=width * ((window_size * 2) + 1)), Mish(
LayerNorm(width), nO=width,
nI=width * ((window_size * 2) + 1),
dropout=0.0,
normalize=True
),
) )
model = clone(residual(cnn), depth) model = clone(residual(cnn), depth)
model.set_dim("nO", width) model.set_dim("nO", width)