2016-11-13 15:45:21 +03:00
|
|
|
from __future__ import print_function
|
|
|
|
from __future__ import unicode_literals
|
|
|
|
from __future__ import division
|
|
|
|
|
|
|
|
import pathlib
|
|
|
|
import plac
|
|
|
|
import random
|
|
|
|
from collections import Counter
|
|
|
|
import numpy as np
|
|
|
|
|
|
|
|
from collections import defaultdict
|
|
|
|
from itertools import count
|
|
|
|
|
2016-11-13 17:48:43 +03:00
|
|
|
#import _gdynet as dynet
|
|
|
|
#from _gdynet import cg
|
2016-11-13 15:45:21 +03:00
|
|
|
import dynet
|
|
|
|
from dynet import cg
|
|
|
|
|
|
|
|
|
|
|
|
class Vocab:
|
|
|
|
def __init__(self, w2i=None):
|
|
|
|
if w2i is None: w2i = defaultdict(count(0).next)
|
|
|
|
self.w2i = dict(w2i)
|
|
|
|
self.i2w = {i:w for w,i in w2i.iteritems()}
|
|
|
|
@classmethod
|
|
|
|
def from_corpus(cls, corpus):
|
|
|
|
w2i = defaultdict(count(0).next)
|
|
|
|
for sent in corpus:
|
|
|
|
[w2i[word] for word in sent]
|
|
|
|
return Vocab(w2i)
|
|
|
|
|
|
|
|
def size(self):
|
|
|
|
return len(self.w2i.keys())
|
|
|
|
|
|
|
|
|
|
|
|
def read_data(path):
|
|
|
|
with path.open() as file_:
|
|
|
|
sent = []
|
|
|
|
for line in file_:
|
|
|
|
line = line.strip().split()
|
|
|
|
if not line:
|
|
|
|
if sent:
|
|
|
|
yield sent
|
|
|
|
sent = []
|
|
|
|
else:
|
|
|
|
pieces = line
|
|
|
|
w = pieces[1]
|
|
|
|
pos = pieces[3]
|
|
|
|
sent.append((w, pos))
|
|
|
|
|
|
|
|
|
|
|
|
def get_vocab(train, test):
|
|
|
|
words = []
|
|
|
|
tags = []
|
|
|
|
wc = Counter()
|
|
|
|
for s in train:
|
|
|
|
for w, p in s:
|
|
|
|
words.append(w)
|
|
|
|
tags.append(p)
|
|
|
|
wc[w] += 1
|
|
|
|
words.append("_UNK_")
|
|
|
|
#words=[w if wc[w] > 1 else "_UNK_" for w in words]
|
|
|
|
tags.append("_START_")
|
|
|
|
|
|
|
|
for s in test:
|
|
|
|
for w, p in s:
|
|
|
|
words.append(w)
|
|
|
|
vw = Vocab.from_corpus([words])
|
|
|
|
vt = Vocab.from_corpus([tags])
|
|
|
|
return words, tags, wc, vw, vt
|
|
|
|
|
|
|
|
|
|
|
|
class BiTagger(object):
|
2016-11-13 17:48:43 +03:00
|
|
|
def __init__(self, vw, vt, nwords, ntags):
|
|
|
|
self.vw = vw
|
|
|
|
self.vt = vt
|
2016-11-13 15:45:21 +03:00
|
|
|
self.nwords = nwords
|
|
|
|
self.ntags = ntags
|
|
|
|
|
2016-11-13 17:48:43 +03:00
|
|
|
self.UNK = self.vw.w2i["_UNK_"]
|
|
|
|
|
2016-11-13 15:45:21 +03:00
|
|
|
self._model = dynet.Model()
|
|
|
|
self._sgd = dynet.SimpleSGDTrainer(self._model)
|
|
|
|
|
|
|
|
self._E = self._model.add_lookup_parameters((self.nwords, 128))
|
|
|
|
self._p_t1 = self._model.add_lookup_parameters((self.ntags, 30))
|
|
|
|
|
|
|
|
self._pH = self._model.add_parameters((32, 50*2))
|
|
|
|
self._pO = self._model.add_parameters((self.ntags, 32))
|
|
|
|
|
|
|
|
self._fwd_lstm = dynet.LSTMBuilder(1, 128, 50, self._model)
|
|
|
|
self._bwd_lstm = dynet.LSTMBuilder(1, 128, 50, self._model)
|
2016-11-13 17:48:43 +03:00
|
|
|
self._words_batch = []
|
|
|
|
self._tags_batch = []
|
|
|
|
self._minibatch_size = 32
|
2016-11-13 15:45:21 +03:00
|
|
|
|
2016-11-13 17:48:43 +03:00
|
|
|
def __call__(self, words):
|
2016-11-13 15:45:21 +03:00
|
|
|
dynet.renew_cg()
|
2016-11-13 17:48:43 +03:00
|
|
|
word_ids = [self.vw.w2i.get(w, self.UNK) for w in words]
|
|
|
|
wembs = [self._E[w] for w in word_ids]
|
2016-11-13 15:45:21 +03:00
|
|
|
|
|
|
|
f_state = self._fwd_lstm.initial_state()
|
|
|
|
b_state = self._bwd_lstm.initial_state()
|
|
|
|
|
|
|
|
fw = [x.output() for x in f_state.add_inputs(wembs)]
|
|
|
|
bw = [x.output() for x in b_state.add_inputs(reversed(wembs))]
|
|
|
|
|
|
|
|
H = dynet.parameter(self._pH)
|
|
|
|
O = dynet.parameter(self._pO)
|
|
|
|
|
2016-11-13 17:48:43 +03:00
|
|
|
tags = []
|
2016-11-13 15:45:21 +03:00
|
|
|
for i, (f, b) in enumerate(zip(fw, reversed(bw))):
|
|
|
|
r_t = O * (dynet.tanh(H * dynet.concatenate([f, b])))
|
|
|
|
out = dynet.softmax(r_t)
|
2016-11-13 17:48:43 +03:00
|
|
|
tags.append(self.vt.i2w[np.argmax(out.npvalue())])
|
|
|
|
return tags
|
|
|
|
|
|
|
|
def update(self, words, tags):
|
|
|
|
self._words_batch.append(words)
|
|
|
|
self._tags_batch.append(tags)
|
|
|
|
if len(self._words_batch) == self._minibatch_size:
|
|
|
|
loss = self.update_batch(self._words_batch, self._tags_batch)
|
|
|
|
self._words_batch = []
|
|
|
|
self._tags_batch = []
|
|
|
|
else:
|
|
|
|
loss = 0
|
|
|
|
return loss
|
|
|
|
|
|
|
|
def update_batch(self, words_batch, tags_batch):
|
2016-11-13 15:45:21 +03:00
|
|
|
dynet.renew_cg()
|
2016-11-13 17:48:43 +03:00
|
|
|
length = max(len(words) for words in words_batch)
|
|
|
|
word_ids = np.zeros((length, len(words_batch)), dtype='int32')
|
|
|
|
for j, words in enumerate(words_batch):
|
|
|
|
for i, word in enumerate(words):
|
|
|
|
word_ids[i, j] = self.vw.w2i.get(word, self.UNK)
|
|
|
|
tag_ids = np.zeros((length, len(words_batch)), dtype='int32')
|
|
|
|
for j, tags in enumerate(tags_batch):
|
|
|
|
for i, tag in enumerate(tags):
|
|
|
|
tag_ids[i, j] = self.vt.w2i.get(tag, self.UNK)
|
|
|
|
wembs = [dynet.lookup_batch(self._E, word_ids[i]) for i in range(length)]
|
2016-11-13 15:45:21 +03:00
|
|
|
wembs = [dynet.noise(we, 0.1) for we in wembs]
|
|
|
|
|
|
|
|
f_state = self._fwd_lstm.initial_state()
|
|
|
|
b_state = self._bwd_lstm.initial_state()
|
|
|
|
|
|
|
|
fw = [x.output() for x in f_state.add_inputs(wembs)]
|
|
|
|
bw = [x.output() for x in b_state.add_inputs(reversed(wembs))]
|
|
|
|
|
|
|
|
H = dynet.parameter(self._pH)
|
|
|
|
O = dynet.parameter(self._pO)
|
|
|
|
|
|
|
|
errs = []
|
2016-11-13 17:48:43 +03:00
|
|
|
for i, (f, b) in enumerate(zip(fw, reversed(bw))):
|
2016-11-13 15:45:21 +03:00
|
|
|
f_b = dynet.concatenate([f,b])
|
|
|
|
r_t = O * (dynet.tanh(H * f_b))
|
2016-11-13 17:48:43 +03:00
|
|
|
err = dynet.pickneglogsoftmax_batch(r_t, tag_ids[i])
|
|
|
|
errs.append(dynet.sum_batches(err))
|
2016-11-13 15:45:21 +03:00
|
|
|
sum_errs = dynet.esum(errs)
|
|
|
|
squared = -sum_errs # * sum_errs
|
2016-11-13 17:48:43 +03:00
|
|
|
losses = sum_errs.scalar_value()
|
2016-11-13 15:45:21 +03:00
|
|
|
sum_errs.backward()
|
2016-11-13 17:48:43 +03:00
|
|
|
self._sgd.update()
|
|
|
|
return losses
|
2016-11-13 15:45:21 +03:00
|
|
|
|
|
|
|
|
|
|
|
def main(train_loc, dev_loc, model_dir):
|
|
|
|
train_loc = pathlib.Path(train_loc)
|
|
|
|
dev_loc = pathlib.Path(dev_loc)
|
|
|
|
|
|
|
|
train = list(read_data((train_loc)))
|
|
|
|
test = list(read_data(dev_loc))
|
|
|
|
|
2016-11-13 17:48:43 +03:00
|
|
|
words, tags, wc, vw, vt = get_vocab(train, test)
|
2016-11-13 15:45:21 +03:00
|
|
|
|
|
|
|
UNK = vw.w2i["_UNK_"]
|
|
|
|
nwords = vw.size()
|
|
|
|
ntags = vt.size()
|
|
|
|
|
2016-11-13 17:48:43 +03:00
|
|
|
tagger = BiTagger(vw, vt, nwords, ntags)
|
2016-11-13 15:45:21 +03:00
|
|
|
|
|
|
|
tagged = loss = 0
|
2016-11-13 17:48:43 +03:00
|
|
|
|
2016-11-13 15:45:21 +03:00
|
|
|
for ITER in xrange(50):
|
|
|
|
random.shuffle(train)
|
|
|
|
for i, s in enumerate(train,1):
|
|
|
|
if i % 5000 == 0:
|
2016-11-13 17:48:43 +03:00
|
|
|
tagger._sgd.status()
|
2016-11-13 15:45:21 +03:00
|
|
|
print(loss / tagged)
|
|
|
|
loss = 0
|
|
|
|
tagged = 0
|
|
|
|
if i % 10000 == 0:
|
|
|
|
good = bad = 0.0
|
|
|
|
for sent in test:
|
2016-11-13 17:48:43 +03:00
|
|
|
#word_ids = [vw.w2i.get(w, UNK) for w, t in sent]
|
|
|
|
tags = tagger([w for w, t in sent])
|
2016-11-13 15:45:21 +03:00
|
|
|
golds = [t for w, t in sent]
|
|
|
|
for go, gu in zip(golds, tags):
|
|
|
|
if go == gu:
|
|
|
|
good += 1
|
|
|
|
else:
|
|
|
|
bad += 1
|
|
|
|
print(good / (good+bad))
|
2016-11-13 17:48:43 +03:00
|
|
|
loss += tagger.update([w for w, t in s], [t for w, t in s])
|
|
|
|
tagged += len(s)
|
2016-11-13 15:45:21 +03:00
|
|
|
|
|
|
|
|
|
|
|
if __name__ == '__main__':
|
|
|
|
plac.call(main)
|