forked from JViolante/sentence-classification
/
sentence-regression.py
323 lines (303 loc) · 16.6 KB
/
sentence-regression.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
312
313
314
315
316
317
318
319
320
321
322
323
import numpy as np
import theano
import math
import csv
import keras
import sklearn
import gensim
import random
import scipy
import keras
from keras.preprocessing import sequence
from keras.preprocessing.text import Tokenizer
from keras.models import Sequential , Graph
from keras.layers.core import Dense , Dropout , Activation , Merge , Flatten
from keras.layers.convolutional import Convolution1D, MaxPooling1D
from keras.layers import Embedding , LSTM
from sklearn.base import BaseEstimator
from sklearn.kernel_ridge import KernelRidge
from gensim.models.word2vec import Word2Vec
from gensim.models.doc2vec import Doc2Vec , TaggedDocument
from word_movers_knn import WordMoversKNN
from sklearn import preprocessing
from sklearn.metrics import mean_squared_error , mean_absolute_error
from geopy import distance
# size of the word embeddings
embeddings_dim = 300
# maximum number of words to consider in the representations
max_features = 20000
# maximum length of a sentence
max_sent_len = 50
# percentage of the data used for model training
percent = 0.75
# special case for geocoding problems -- regression in which the metric to optimize is the geospatial distance instead of RMSE
is_geocoding = True
#number of dimensions in regression problem
reg_dimensions = 2
def geodistance( coords1 , coords2 ):
lat1 , lon1 = coords1[ : 2]
lat2 , lon2 = coords2[ : 2]
try: return distance.vincenty( ( lat1 , lon1 ) , ( lat2 , lon2 ) ).meters / 1000.0
except: return distance.great_circle( ( lat1 , lon1 ) , ( lat2 , lon2 ) ).meters / 1000.0
def geoloss( a , b ):
aa = theano.tensor.deg2rad( a )
bb = theano.tensor.deg2rad( b )
sin_lat1 = theano.tensor.sin( aa[:,0] )
cos_lat1 = theano.tensor.cos( aa[:,0] )
sin_lat2 = theano.tensor.sin( bb[:,0] )
cos_lat2 = theano.tensor.cos( bb[:,0] )
delta_lng = bb[:,1] - aa[:,1]
cos_delta_lng = theano.tensor.cos(delta_lng)
sin_delta_lng = theano.tensor.sin(delta_lng)
d = theano.tensor.arctan2(theano.tensor.sqrt((cos_lat2 * sin_delta_lng) ** 2 + (cos_lat1 * sin_lat2 - sin_lat1 * cos_lat2 * cos_delta_lng) ** 2), sin_lat1 * sin_lat2 + cos_lat1 * cos_lat2 * cos_delta_lng )
return theano.tensor.mean( 6371.0088 * d , axis = -1 )
print ("")
print ("Reading pre-trained word embeddings...")
embeddings = dict( )
embeddings = Word2Vec.load_word2vec_format( "GoogleNews-vectors-negative300.bin.gz" , binary=True )
print ("Reading text data for regression and building representations...")
data = [ ( row["message"] , ( float( row["latitude"] ) , float( row["longitude"] ) ) ) for row in csv.DictReader(open("test-data-geo.txt"), delimiter='\t', quoting=csv.QUOTE_NONE) ]
is_geocoding = True
reg_dimensions = 2
#data = [ ( row["sentence"] , float( row["rating"] ) ) for row in csv.DictReader(open("test-data-rated.txt"), delimiter='\t', quoting=csv.QUOTE_NONE) ]
#is_geocoding = False
#reg_dimensions = 1
random.shuffle( data )
train_size = int(len(data) * percent)
train_texts = [ txt for ( txt, label ) in data[0:train_size] ]
test_texts = [ txt for ( txt, label ) in data[train_size:-1] ]
train_labels = [ label for ( txt , label ) in data[0:train_size] ]
test_labels = [ label for ( txt , label ) in data[train_size:-1] ]
tokenizer = Tokenizer(nb_words=max_features, filters=keras.preprocessing.text.base_filter(), lower=True, split=" ")
tokenizer.fit_on_texts(train_texts)
train_sequences = sequence.pad_sequences( tokenizer.texts_to_sequences( train_texts ) , maxlen=max_sent_len )
test_sequences = sequence.pad_sequences( tokenizer.texts_to_sequences( test_texts ) , maxlen=max_sent_len )
train_matrix = tokenizer.texts_to_matrix( train_texts )
test_matrix = tokenizer.texts_to_matrix( test_texts )
embedding_weights = np.zeros( ( max_features , embeddings_dim ) )
for word,index in tokenizer.word_index.items():
if index < max_features:
try: embedding_weights[index,:] = embeddings[word]
except: embedding_weights[index,:] = np.random.rand( 1 , embeddings_dim )
print ("")
print ("Method = Linear ridge regression with bag-of-words features")
model = KernelRidge( kernel='linear' )
model.fit( train_matrix , train_labels )
results = model.predict( test_matrix )
if not(is_geocoding):
print ("RMSE = " + repr( np.sqrt(mean_squared_error( test_labels , results )) ) )
print ("MAE = " + repr( mean_absolute_error( test_labels , results ) ) )
else:
print ("Mean error = " + repr( np.mean( [ geodistance( results[i] , test_labels[i] ) for i in range(results.shape[0]) ] ) ) )
print ("Median error = " + repr( np.median( [ geodistance( results[i] , test_labels[i] ) for i in range(results.shape[0]) ] ) ) )
print ("")
print ("Method = MLP with bag-of-words features")
np.random.seed(0)
model = Sequential()
model.add(Dense(embeddings_dim, input_dim=train_matrix.shape[1], init='uniform', activation='relu'))
model.add(Dropout(0.25))
model.add(Dense(embeddings_dim, activation='relu'))
model.add(Dropout(0.25))
model.add(Dense(reg_dimensions, activation='sigmoid'))
if not(is_geocoding): model.compile(loss='mean_absolute_error', optimizer='adam')
else: model.compile(loss=geoloss, optimizer='adam')
model.fit( train_matrix , train_labels , nb_epoch=30, batch_size=16)
results = model.predict( test_matrix )
if not(is_geocoding):
print ("RMSE = " + repr( np.sqrt(mean_squared_error( test_labels , results )) ) )
print ("MAE = " + repr( mean_absolute_error( test_labels , results ) ) )
else:
print ("Mean error = " + repr( np.mean( [ geodistance( results[i] , test_labels[i] ) for i in range(results.shape[0]) ] ) ) )
print ("Median error = " + repr( np.median( [ geodistance( results[i] , test_labels[i] ) for i in range(results.shape[0]) ] ) ) )
print ("")
print ("Method = Stack of two LSTMs")
np.random.seed(0)
model = Sequential()
model.add(Embedding(max_features, embeddings_dim, input_length=max_sent_len, mask_zero=True, weights=[embedding_weights] ))
model.add(Dropout(0.25))
model.add(LSTM(output_dim=embeddings_dim , activation='sigmoid', inner_activation='hard_sigmoid', return_sequences=True))
model.add(Dropout(0.25))
model.add(LSTM(output_dim=embeddings_dim , activation='sigmoid', inner_activation='hard_sigmoid'))
model.add(Dropout(0.25))
model.add(Dense(reg_dimensions))
model.add(Activation('sigmoid'))
if not(is_geocoding): model.compile(loss='mean_absolute_error', optimizer='adam')
else: model.compile(loss=geoloss, optimizer='adam')
model.fit( train_sequences , train_labels , nb_epoch=30, batch_size=16)
results = model.predict( test_sequences )
if not(is_geocoding):
print ("RMSE = " + repr( np.sqrt(mean_squared_error( test_labels , results )) ) )
print ("MAE = " + repr( mean_absolute_error( test_labels , results ) ) )
else:
print ("Mean error = " + repr( np.mean( [ geodistance( results[i] , test_labels[i] ) for i in range(results.shape[0]) ] ) ) )
print ("Median error = " + repr( np.median( [ geodistance( results[i] , test_labels[i] ) for i in range(results.shape[0]) ] ) ) )
print ("")
print ("Method = CNN from the paper 'Convolutional Neural Networks for Sentence Classification'")
np.random.seed(0)
nb_filter = embeddings_dim
model = Graph()
model.add_input(name='input', input_shape=(max_sent_len,), dtype=int)
model.add_node(Embedding(max_features, embeddings_dim, input_length=max_sent_len, mask_zero=False, weights=[embedding_weights] ), name='embedding', input='input')
model.add_node(Dropout(0.25), name='dropout_embedding', input='embedding')
for n_gram in [3, 5, 7]:
model.add_node(Convolution1D(nb_filter=nb_filter, filter_length=n_gram, border_mode='valid', activation='relu', subsample_length=1, input_dim=embeddings_dim, input_length=max_sent_len), name='conv_' + str(n_gram), input='dropout_embedding')
model.add_node(MaxPooling1D(pool_length=max_sent_len - n_gram + 1), name='maxpool_' + str(n_gram), input='conv_' + str(n_gram))
model.add_node(Flatten(), name='flat_' + str(n_gram), input='maxpool_' + str(n_gram))
model.add_node(Dropout(0.25), name='dropout', inputs=['flat_' + str(n) for n in [3, 5, 7]])
model.add_node(Dense(reg_dimensions, input_dim=nb_filter * len([3, 5, 7])), name='dense', input='dropout')
model.add_node(Activation('sigmoid'), name='sigmoid', input='dense')
model.add_output(name='output', input='sigmoid')
if not(is_geocoding): model.compile(loss={'output': 'mean_absolute_error'}, optimizer='adam')
else: model.compile(loss={'output': geoloss}, optimizer='adam')
model.fit({'input': train_sequences, 'output': train_labels}, batch_size=16, nb_epoch=30)
results = np.array(model.predict({'input': test_sequences}, batch_size=16)['output'])
if not(is_geocoding):
print ("RMSE = " + repr( np.sqrt(mean_squared_error( test_labels , results )) ) )
print ("MAE = " + repr( mean_absolute_error( test_labels , results ) ) )
else:
print ("Mean error = " + repr( np.mean( [ geodistance( results[i] , test_labels[i] ) for i in range(results.shape[0]) ] ) ) )
print ("Median error = " + repr( np.median( [ geodistance( results[i] , test_labels[i] ) for i in range(results.shape[0]) ] ) ) )
print ("")
print ("Method = Bidirectional LSTM")
np.random.seed(0)
model = Graph()
model.add_input(name='input', input_shape=(max_sent_len,), dtype=int)
model.add_node(Embedding( max_features, embeddings_dim, input_length=max_sent_len, mask_zero=True, weights=[embedding_weights] ), name='embedding', input='input')
model.add_node(LSTM(embeddings_dim, activation='sigmoid', inner_activation='hard_sigmoid', return_sequences=True), name='forward1', input='embedding')
model.add_node(Dropout(0.25), name="dropout1", input='forward1')
model.add_node(LSTM(embeddings_dim, activation='sigmoid', inner_activation='hard_sigmoid'), name='forward2', input='forward1')
model.add_node(LSTM(embeddings_dim, activation='sigmoid', inner_activation='hard_sigmoid', go_backwards=True, return_sequences=True), name='backward1', input='embedding')
model.add_node(Dropout(0.25), name="dropout2", input='backward1')
model.add_node(LSTM(embeddings_dim, activation='sigmoid', inner_activation='hard_sigmoid', go_backwards=True), name='backward2', input='backward1')
model.add_node(Dropout(0.25), name='dropout', inputs=['forward2', 'backward2'])
model.add_node(Dense(reg_dimensions, activation='sigmoid'), name='sigmoid', input='dropout')
model.add_output(name='output', input='sigmoid')
if not(is_geocoding): model.compile(loss={'output': 'mean_absolute_error'}, optimizer='adam')
else: model.compile(loss={'output': geoloss}, optimizer='adam')
model.fit({'input': train_sequences, 'output': train_labels}, batch_size=16, nb_epoch=30)
results = np.array(model.predict({'input': test_sequences}, batch_size=16)['output'])
if not(is_geocoding):
print ("RMSE = " + repr( np.sqrt(mean_squared_error( test_labels , results )) ) )
print ("MAE = " + repr( mean_absolute_error( test_labels , results ) ) )
else:
print ("Mean error = " + repr( np.mean( [ geodistance( results[i] , test_labels[i] ) for i in range(results.shape[0]) ] ) ) )
print ("Median error = " + repr( np.median( [ geodistance( results[i] , test_labels[i] ) for i in range(results.shape[0]) ] ) ) )
print ("")
print ("Method = CNN-LSTM")
np.random.seed(0)
filter_length = 3
nb_filter = embeddings_dim
pool_length = 2
model = Sequential()
model.add(Embedding(max_features, embeddings_dim, input_length=max_sent_len, weights=[embedding_weights]))
model.add(Dropout(0.25))
model.add(Convolution1D(nb_filter=nb_filter, filter_length=filter_length, border_mode='valid', activation='relu', subsample_length=1))
model.add(MaxPooling1D(pool_length=pool_length))
model.add(LSTM(embeddings_dim))
model.add(Dense(reg_dimensions))
model.add(Activation('sigmoid'))
if not(is_geocoding): model.compile(loss='mean_absolute_error', optimizer='adam')
else: model.compile(loss=geoloss, optimizer='adam')
model.fit( train_sequences , train_labels , nb_epoch=30, batch_size=16)
results = model.predict( test_sequences )
if not(is_geocoding):
print ("RMSE = " + repr( np.sqrt(mean_squared_error( test_labels , results )) ) )
print ("MAE = " + repr( mean_absolute_error( test_labels , results ) ) )
else:
print ("Mean error = " + repr( np.mean( [ geodistance( results[i] , test_labels[i] ) for i in range(results.shape[0]) ] ) ) )
print ("Median error = " + repr( np.median( [ geodistance( results[i] , test_labels[i] ) for i in range(results.shape[0]) ] ) ) )
print ("")
print ("Method = Linear ridge regression with doc2vec features")
np.random.seed(0)
class LabeledLineSentence(object):
def __init__(self, data ): self.data = data
def __iter__(self):
for uid, line in enumerate( self.data ): yield TaggedDocument( line.split(" ") , ["S_%s" % uid] )
model = Doc2Vec( alpha=0.025 , min_alpha=0.025 )
sentences = LabeledLineSentence( train_texts + test_texts )
model.build_vocab( sentences )
model.train( sentences )
for w in model.vocab.keys():
try : model[w] = embeddings[w]
except : continue
for epoch in range(10):
model.train(sentences)
model.alpha -= 0.002
model.min_alpha = model.alpha
train_rep = np.array( [ model.docvecs[i] for i in range( train_matrix.shape[0] ) ] )
test_rep = np.array( [ model.docvecs[i + train_matrix.shape[0]] for i in range( test_matrix.shape[0] ) ] )
model = KernelRidge( kernel='linear' )
model.fit( train_rep , train_labels )
results = model.predict( test_rep )
if not(is_geocoding):
print ("RMSE = " + repr( np.sqrt(mean_squared_error( test_labels , results )) ) )
print ("MAE = " + repr( mean_absolute_error( test_labels , results ) ) )
else:
print ("Mean error = " + repr( np.mean( [ geodistance( results[i] , test_labels[i] ) for i in range(results.shape[0]) ] ) ) )
print ("Median error = " + repr( np.median( [ geodistance( results[i] , test_labels[i] ) for i in range(results.shape[0]) ] ) ) )
print ("")
print ("Method = Kernel ridge regression with doc2vec features")
np.random.seed(0)
class LabeledLineSentence(object):
def __init__(self, data ): self.data = data
def __iter__(self):
for uid, line in enumerate( self.data ): yield TaggedDocument( line.split(" ") , ["S_%s" % uid] )
model = Doc2Vec( alpha=0.025 , min_alpha=0.025 )
sentences = LabeledLineSentence( train_texts + test_texts )
model.build_vocab( sentences )
model.train( sentences )
for w in model.vocab.keys():
try : model[w] = embeddings[w]
except : continue
for epoch in range(10):
model.train(sentences)
model.alpha -= 0.002
model.min_alpha = model.alpha
train_rep = np.array( [ model.docvecs[i] for i in range( train_matrix.shape[0] ) ] )
test_rep = np.array( [ model.docvecs[i + train_matrix.shape[0]] for i in range( test_matrix.shape[0] ) ] )
model = KernelRidge( kernel='rbf' )
model.fit( train_rep , train_labels )
results = model.predict( test_rep )
if not(is_geocoding):
print ("RMSE = " + repr( np.sqrt(mean_squared_error( test_labels , results )) ) )
print ("MAE = " + repr( mean_absolute_error( test_labels , results ) ) )
else:
print ("Mean error = " + repr( np.mean( [ geodistance( results[i] , test_labels[i] ) for i in range(results.shape[0]) ] ) ) )
print ("Median error = " + repr( np.median( [ geodistance( results[i] , test_labels[i] ) for i in range(results.shape[0]) ] ) ) )
print ("")
print ("Method = MLP with doc2vec features")
np.random.seed(0)
class LabeledLineSentence(object):
def __init__(self, data ): self.data = data
def __iter__(self):
for uid, line in enumerate( self.data ): yield TaggedDocument( line.split(" ") , ["S_%s" % uid] )
model = Doc2Vec( alpha=0.025 , min_alpha=0.025 )
sentences = train_texts + test_texts
sentences = LabeledLineSentence( sentences )
model.build_vocab( sentences )
model.train( sentences )
for w in model.vocab.keys():
try : model[w] = embeddings[w]
except : continue
for epoch in range(10):
model.train(sentences)
model.alpha -= 0.002
model.min_alpha = model.alpha
train_rep = np.array( [ model.docvecs[i] for i in range( train_matrix.shape[0] ) ] )
test_rep = np.array( [ model.docvecs[i + train_matrix.shape[0]] for i in range( test_matrix.shape[0] ) ] )
model = Sequential()
model.add(Dense(embeddings_dim, input_dim=train_rep.shape[1], init='uniform', activation='relu'))
model.add(Dropout(0.25))
model.add(Dense(embeddings_dim, activation='relu'))
model.add(Dropout(0.25))
model.add(Dense(reg_dimensions, activation='sigmoid'))
if not(is_geocoding): model.compile(loss='mean_absolute_error', optimizer='adam')
else: model.compile(loss=geoloss, optimizer='adam')
model.fit( train_rep , train_labels , nb_epoch=30, batch_size=16)
results = model.predict( test_rep )
if not(is_geocoding):
print ("RMSE = " + repr( np.sqrt(mean_squared_error( test_labels , results )) ) )
print ("MAE = " + repr( mean_absolute_error( test_labels , results ) ) )
else:
print ("Mean error = " + repr( np.mean( [ geodistance( results[i] , test_labels[i] ) for i in range(results.shape[0]) ] ) ) )
print ("Median error = " + repr( np.median( [ geodistance( results[i] , test_labels[i] ) for i in range(results.shape[0]) ] ) ) )