-
Notifications
You must be signed in to change notification settings - Fork 1
/
deep_qa.py
311 lines (236 loc) · 10.2 KB
/
deep_qa.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
import pdb
import operator
import cPickle
import numpy as np
import pandas as pd
#import sklearn
#from sklearn.feature_extraction.text import CountVectorizer
import nltk
try:
import cochranenlp
from cochranenlp.readers.biviewer import PDFBiViewer
except:
print("cochrannlp not found!")
import keras
from keras.layers.embeddings import Embedding
from keras.layers.core import Dense, Merge, Activation, RepeatVector, TimeDistributedDense, Dropout
from keras.layers import recurrent
from keras.models import Sequential, model_from_json
from keras.preprocessing.sequence import pad_sequences
from keras.preprocessing.text import text_to_word_sequence
from keras.preprocessing.text import Tokenizer
from keras.layers.recurrent import LSTM, GRU
import gensim
from gensim.models import Word2Vec
START_STR = "startstartstart"
STOP_STR = "stopstopstop"
'''
def _get_init_vectors(vectorizer, wv, unknown_words_to_vecs):
init_vectors = []
for token_idx, t in enumerate(vectorizer.vocabulary):
try:
init_vectors.append(wv[t])
except:
init_vectors.append(unknown_words_to_vecs[t])
init_vectors = np.vstack(init_vectors)
return init_vectors
'''
def load_trained_w2v_model(path="/Users/byron/dev/Deep-PICO/PubMed-w2v.bin"):
m = Word2Vec.load_word2vec_format(path, binary=True)
return m
def get_docs_and_intervention_summaries(pico_elem_str="CHAR_INTERVENTIONS"):
pairs = []
p = PDFBiViewer()
for study in p:
cdsr_entry = study.cochrane
text = study.studypdf['text']
intervention_text = cdsr_entry["CHARACTERISTICS"][pico_elem_str]
if intervention_text is not None:
#pairs.append((nltk.word_tokenize(text),
# nltk.word_tokenize(intervention_text)))
pairs.append((text_to_word_sequence(text),
text_to_word_sequence(intervention_text)))
return pairs
class ISummarizer:
# 100000
def __init__(self, pairs, nb_words=10000, hidden_size=512, max_input_size=3000, max_output_size=15):
self.pairs = pairs
self.nb_words = nb_words + 2 # number of words; +2 for start and stop tokens!
self.max_input_size = max_input_size
self.max_output_size = max_output_size + 2 # again +2 for start/stop
self.hidden_size = hidden_size
print("loading pre-trained word vectors...")
self.wv = load_trained_w2v_model()
# here you want to add start and stop
print("OK!")
self.word_embedding_size = self.wv.vector_size
# call to sequences
# call init_word_vectors
print("building sequences...")
self.build_sequences()
print("initializing word vectors...")
self.init_word_vectors()
print("ok!")
def build_sequences(self):
self.tokenizer = Tokenizer(nb_words=self.nb_words)
self.raw_input_texts = [START_STR + " " + " ".join(pair[0]) + " " + STOP_STR for pair in self.pairs]
self.raw_output_texts = [START_STR + " " + " ".join(pair[1]) + " " + STOP_STR for pair in self.pairs]
def _get_max(seqs):
return max([len(seq) for seq in seqs])
self.tokenizer.fit_on_texts(self.raw_input_texts+self.raw_output_texts)
self.word_indices_to_words = {}
for token, idx in self.tokenizer.word_index.items():
self.word_indices_to_words[idx] = token
self.input_sequences = list(self.tokenizer.texts_to_sequences_generator(self.raw_input_texts))
#self.max_input_len = _get_max(self.input_sequences)
#X_train = sequence.pad_sequences(X_train, maxlen=maxlen)
#X_test = sequence.pad_sequences(X_test, maxlen=maxlen)
self.input_sequences = list(pad_sequences(self.input_sequences, maxlen=self.max_input_size))
self.output_sequences = list(self.tokenizer.texts_to_sequences_generator(self.raw_output_texts))
self.output_sequences = list(pad_sequences(self.output_sequences, maxlen=self.max_output_size))
def init_word_vectors(self):
self.init_vectors = []
unknown_words_to_vecs = {}
for t, token_idx in self.tokenizer.word_index.items():
if token_idx <= self.nb_words:
try:
self.init_vectors.append(self.wv[t])
except:
if t not in unknown_words_to_vecs:
# randomly initialize
unknown_words_to_vecs[t] = np.random.random(
self.word_embedding_size)*-2 + 1
self.init_vectors.append(unknown_words_to_vecs[t])
self.init_vectors = np.vstack(self.init_vectors)
def build_model(self):
self.model = Sequential()
self.model.add(Embedding(self.nb_words, self.word_embedding_size, weights=[self.init_vectors]))
###
# run embeddings through a Gated Recurrent Unit
self.model.add(GRU(self.hidden_size))
#self.model.add(Dropout(0.1))
self.model.add(Dense(self.hidden_size))
self.model.add(Activation('relu'))
self.model.add(RepeatVector(self.max_output_size))
self.model.add(GRU(self.hidden_size, return_sequences=True))
self.model.add(Dropout(0.1))
self.model.add(TimeDistributedDense(self.nb_words, activation="softmax"))
# does cross entropy make sense here?
self.model.compile(loss="categorical_crossentropy", optimizer='adam')
return self.model
def X_y(self):
self.X = np.array(self.input_sequences) # np.zeros((n, self.max_input_size, self.nb_words), dtype=np.bool)
self.Y = np.zeros((len(self.output_sequences), self.max_output_size, self.nb_words), dtype=np.bool)
for i in range(self.X.shape[0]):
#for j, token_idx in enumerate(self.input_sequences[i]):
# self.X[i, j, token_idx] = 1
for j, token_idx in enumerate(self.output_sequences[i]):
self.Y[i, j, token_idx] = 1
print "X shape: %s; Y shape: %s" % (self.X.shape, self.Y.shape)
def decode(self, pred):
text = []
for token_preds in pred:
### it keeps predicting zeros! zeros are for the padding...
cur_pred_index = np.argmax(token_preds) #+ 1 # the tokenizer seems to do 1-indexing!
if cur_pred_index == 0:
text.append("<pad>")
else:
text.append(self.word_indices_to_words[cur_pred_index])
return text
def train(self):
# @TODO revisit; batchsize, etc
print "fitting model..."
self.model.fit(self.X, self.Y)
def all_systems_go():
from keras.callbacks import ModelCheckpoint
pairs = cPickle.load(open("pairs.pickle"))
IS = ISummarizer(pairs)
model = IS.build_model()
IS.X_y()
#print("dumping summarizer!")
#with open("IS.pickle", 'w') as outf:
# cPickle.dump(IS, outf)
# dump the model architecture!
json_string = model.to_json()
open('model_architecture.json', 'w').write(json_string)
print("dumped model!")
n_train = 14500
#X_train, Y_train = IS.X[:n_train], IS.Y[:n_train]
#X_test, Y_test = IS.X[n_train:], IS.Y[n_train:]
# validation_data=(X_test, Y_test),
print "ok... fitting ..."
checkpointer = ModelCheckpoint(filepath="weights.hdf5", verbose=1, save_best_only=True)
model.fit(IS.X[:n_train], IS.Y[:n_train], batch_size=128, nb_epoch=20, verbose=2, callbacks=[checkpointer])
#show_accuracy=True,
#validation_split = .1,
#callbacks=[checkpointer])
def predict():
m = model_from_json(open("model_architecture.json").read())
m.load_weights('weights.hdf5')
pairs = cPickle.load(open("pairs.pickle"))
IS = deep_qa.ISummarizer(pairs)
IS.X_y()
preds = m.predict(IS.X[:1]) # predictions for first example
'''
from keras.callbacks import ModelCheckpoint
model = Sequential()
model.add(Dense(10, input_dim=784, init='uniform'))
model.add(Activation('softmax'))
model.compile(loss='categorical_crossentropy', optimizer='rmsprop')
#saves the model weights after each epoch if the validation loss decreased
checkpointer = ModelCheckpoint(filepath="/tmp/weights.hdf5", verbose=1, save_best_only=True)
model.fit(X_train, Y_train, batch_size=128, nb_epoch=20, verbose=0, validation_data=(X_test, Y_test), callbacks=[checkpointer])
module load cuda
module load python
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/opt/apps/intel14/hdf5/1.8.12/x86_64/lib/
THEANO_FLAGS=mode=FAST_RUN,device=gpu,floatX=float32
'''
def toy():
nb_w = 10
embedding_size = 5
#embeddings = [np.random.random(5) for _ in range(10)]
X = [[9, 8, 3], [4, 5, 1], [3, 6, 3], [6,2,1]]
y = [[1, 2], [3,1], [1,6], [5,2]]
X_arr = np.array(X) #np.zeros((len(X), len(X[0]), 10)) # n sample; time steps; vocab size
y_tens = np.zeros((len(y), len(y[0]), 10)) # n sample; time steps; vocab size
for i in range(len(X)):
#for j, t_idx in enumerate(X[i]):
# X_tens[i, j, t_idx] = 1
for j, t_idx in enumerate(y[i]):
y_tens[i, j, t_idx] = 1
model = Sequential()
model.add(Embedding(10, embedding_size))
max_output_size = 2
hidden_size = 8
###
# I think we only want the output dim here!
#model.add(GRU(self.word_embedding_size, self.hidden_size))
model.add(GRU(hidden_size))
model.add(Dense(hidden_size))
model.add(Activation('relu'))
model.add(RepeatVector(max_output_size))
model.add(GRU(hidden_size, return_sequences=True))
model.add(TimeDistributedDense(nb_w, activation="softmax"))
model.compile(loss='mse', optimizer='adam')
model.fit(X_arr, y_tens)
if __name__ == "__main__":
all_systems_go()
'''
module load cuda
module load python
THEANO_FLAGS=mode=FAST_RUN,device=gpu,floatX=float32
'''
'''
batch_size = 16
embedding_size = 32
hidden_size = 512
print('Build model...')
model = Sequential()
model.add(Embedding(max_features, embedding_size))
model.add(GRU(embedding_size, hidden_size))
model.add(Dense(hidden_size, hidden_size))
model.add(Activation('relu'))
model.add(RepeatVector(maxlen))
model.add(GRU(hidden_size, hidden_size, return_sequences=True))
model.add(TimeDistributedDense(hidden_size, max_features, activation="softmax"))
'''