2017-05-16 11:06:30 +00:00
|
|
|
# coding: utf8
|
|
|
|
from __future__ import unicode_literals
|
2017-05-15 19:46:08 +00:00
|
|
|
from thinc.neural import Model
|
|
|
|
import pytest
|
|
|
|
import numpy
|
|
|
|
|
|
|
|
from ..._ml import chain, Tok2Vec, doc2feats
|
|
|
|
from ...vocab import Vocab
|
|
|
|
from ...pipeline import TokenVectorEncoder
|
|
|
|
from ...syntax.arc_eager import ArcEager
|
|
|
|
from ...syntax.nn_parser import Parser
|
|
|
|
from ...tokens.doc import Doc
|
|
|
|
from ...gold import GoldParse
|
|
|
|
|
|
|
|
|
|
|
|
@pytest.fixture
|
|
|
|
def vocab():
|
|
|
|
return Vocab()
|
|
|
|
|
|
|
|
|
|
|
|
@pytest.fixture
|
|
|
|
def arc_eager(vocab):
|
|
|
|
actions = ArcEager.get_actions(left_labels=['L'], right_labels=['R'])
|
|
|
|
return ArcEager(vocab.strings, actions)
|
|
|
|
|
|
|
|
|
|
|
|
@pytest.fixture
|
|
|
|
def tok2vec():
|
|
|
|
return Tok2Vec(8, 100, preprocess=doc2feats())
|
|
|
|
|
|
|
|
|
|
|
|
@pytest.fixture
|
|
|
|
def parser(vocab, arc_eager):
|
|
|
|
return Parser(vocab, moves=arc_eager, model=None)
|
|
|
|
|
|
|
|
@pytest.fixture
|
|
|
|
def model(arc_eager, tok2vec):
|
2017-07-19 22:16:52 +00:00
|
|
|
return Parser.Model(arc_eager.n_moves, token_vector_width=tok2vec.nO)[0]
|
2017-05-15 19:46:08 +00:00
|
|
|
|
|
|
|
@pytest.fixture
|
|
|
|
def doc(vocab):
|
|
|
|
return Doc(vocab, words=['a', 'b', 'c'])
|
|
|
|
|
|
|
|
@pytest.fixture
|
|
|
|
def gold(doc):
|
|
|
|
return GoldParse(doc, heads=[1, 1, 1], deps=['L', 'ROOT', 'R'])
|
2017-07-19 22:16:52 +00:00
|
|
|
|
|
|
|
|
2017-05-15 19:46:08 +00:00
|
|
|
def test_can_init_nn_parser(parser):
|
|
|
|
assert parser.model is None
|
|
|
|
|
|
|
|
|
2017-05-16 14:17:30 +00:00
|
|
|
def test_build_model(parser):
|
2017-07-19 22:16:52 +00:00
|
|
|
parser.model = Parser.Model(parser.moves.n_moves)[0]
|
2017-05-15 19:46:08 +00:00
|
|
|
assert parser.model is not None
|
|
|
|
|
|
|
|
|
2017-05-16 14:17:30 +00:00
|
|
|
def test_predict_doc(parser, tok2vec, model, doc):
|
2017-07-19 22:16:52 +00:00
|
|
|
doc.tensor = tok2vec([doc])[0]
|
2017-05-15 19:46:08 +00:00
|
|
|
parser.model = model
|
2017-05-19 23:11:29 +00:00
|
|
|
parser(doc)
|
2017-05-15 19:46:08 +00:00
|
|
|
|
|
|
|
|
2017-05-16 14:17:30 +00:00
|
|
|
def test_update_doc(parser, tok2vec, model, doc, gold):
|
2017-05-15 19:46:08 +00:00
|
|
|
parser.model = model
|
2017-05-16 14:17:30 +00:00
|
|
|
tokvecs, bp_tokvecs = tok2vec.begin_update([doc])
|
2017-07-19 22:16:52 +00:00
|
|
|
d_tokvecs = parser.update(([doc], tokvecs), [gold])
|
|
|
|
assert d_tokvecs[0].shape == tokvecs[0].shape
|
2017-05-15 19:46:08 +00:00
|
|
|
def optimize(weights, gradient, key=None):
|
|
|
|
weights -= 0.001 * gradient
|
2017-05-19 23:11:29 +00:00
|
|
|
bp_tokvecs(d_tokvecs, sgd=optimize)
|
2017-07-19 22:16:52 +00:00
|
|
|
assert d_tokvecs[0].sum() == 0.
|
2017-07-20 13:03:10 +00:00
|
|
|
|
|
|
|
|
|
|
|
def test_predict_doc_beam(parser, tok2vec, model, doc):
|
|
|
|
doc.tensor = tok2vec([doc])[0]
|
|
|
|
parser.model = model
|
|
|
|
parser(doc, beam_width=32, beam_density=0.001)
|
|
|
|
for word in doc:
|
|
|
|
print(word.text, word.head, word.dep_)
|
2017-08-18 20:27:42 +00:00
|
|
|
|
|
|
|
|
|
|
|
def test_update_doc_beam(parser, tok2vec, model, doc, gold):
|
|
|
|
parser.model = model
|
|
|
|
tokvecs, bp_tokvecs = tok2vec.begin_update([doc])
|
|
|
|
d_tokvecs = parser.update_beam(([doc], tokvecs), [gold])
|
|
|
|
assert d_tokvecs[0].shape == tokvecs[0].shape
|
|
|
|
def optimize(weights, gradient, key=None):
|
|
|
|
weights -= 0.001 * gradient
|
|
|
|
bp_tokvecs(d_tokvecs, sgd=optimize)
|
|
|
|
assert d_tokvecs[0].sum() == 0.
|
|
|
|
|
|
|
|
|