Blame view
bin/results_cnn.py
12.6 KB
362b552ee
upload system
|
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 |
from keras.models import Sequential
from keras.layers.core import TimeDistributedDense
from keras.layers.advanced_activations import PReLU
from keras.layers.normalization import BatchNormalization
from keras import backend as K
#from keras.layers.core import Dense, Dropout, Activation, Reshape, Flatten
from keras.layers import TimeDistributed, Lambda, Input, merge, Bidirectional
from keras.layers.core import Dense, Activation, Dropout, Flatten
from keras.layers import GlobalAveragePooling1D, GlobalMaxPooling1D, Conv1D, MaxPooling1D
#from keras.layers.convolutional import Convolution1D, MaxPooling1D, AveragePooling1D
from keras.optimizers import SGD, Adadelta, Adam, Adamax, RMSprop
from keras.models import Model
#from keras.layers.recurrent import LSTM, GRU, SimpleRNN
from keras.constraints import maxnorm
from keras.callbacks import Callback, EarlyStopping
from keras.preprocessing import sequence
from keras.layers.embeddings import Embedding
from keras.regularizers import l2, activity_l2
from keras import regularizers
from keras.models import model_from_json
import theano
from theano import tensor
import warnings
import sys
import time
import os
import numpy as np
import fileinput
import math
warnings.filterwarnings("ignore")
def create_class_weight(labels_dict,mu=0.55):
total = np.sum(labels_dict.values())
keys = labels_dict.keys()
class_weight = dict()
for key in keys:
score = math.log(mu*total/float(labels_dict[key]))
class_weight[key] = score if score > 1.0 else 1.0
return class_weight
def score_deft_2017(pred, gold, task):
t = ConfusionMatrix()
for x in range(0, len(gold)):
a = pred[x].tolist()
b = gold[x].tolist()
t.store(a.index(max(a)), b.index(max(b)))
return t.score_deft_2017(task)
def score_semeval_2017(pred, gold):
t = ConfusionMatrix()
for x in range(0, len(gold)):
a = pred[x].tolist()
b = gold[x].tolist()
t.store(a.index(max(a)), b.index(max(b)))
return t.score_semeval_2017()
def score_semeval_2016(pred, gold):
t = ConfusionMatrix()
for x in range(0, len(gold)):
a = pred[x].tolist()
b = gold[x].tolist()
t.store(a.index(max(a)), b.index(max(b)))
return t.score_semeval_2016()
class validation_semeval(Callback):
def __init__(self, dev_files, word_word2vec, maxlen, output, task, patience=5):
super(Callback, self).__init__()
self.best_result = -1.0
self.best_round = 1
self.wait = 0
self.output = output
self.patience = patience
self.counter = 0
self.maxlen = maxlen
self.dev_files = dev_files
self.word_word2vec = word_word2vec
self.task = task
def on_epoch_end(self, epoch, logs={}):
self.counter += 1
current = 0
id_train, X_word_train, y_train = read_sentiment(self.dev_files, self.word_word2vec, self.task)
X_word_train = sequence.pad_sequences(X_word_train, maxlen=self.maxlen, padding='post', truncating='post')
#predict = self.model.predict([X_word_train, X_highlevel_train], batch_size=1024, verbose=1)
predict = self.model.predict(X_word_train, batch_size=1024, verbose=1)
current = score_deft_2017(predict, y_train, self.task)
if self.best_result < current:
self.best_result = current
self.best_round = self.counter
self.wait = 0
json_string = self.model.to_json()
open(self.output+".json", 'w').write(json_string)
self.model.save_weights(self.output+".h5", overwrite=True)
print "
"
print "Dev score 2017 : "+str(current)
print "Best score 2017 : "+str(self.best_result)
print "
"
class ConfusionMatrix(object):
def __init__(self):
self.h = {}
self.total = 0
def store(self, actual, truth):
if actual not in self.h.keys():
self.h[ actual ] = {"tp": 0, "tn": 0, "fp": 0, "fn": 0}
if truth not in self.h.keys():
self.h[ truth ] = {"tp": 0, "tn": 0, "fp": 0, "fn": 0}
if actual == truth:
self.h[actual]["tp"] += 1
else:
self.h[actual]["fp"] += 1
self.h[truth]["fn"] += 1
self.h[truth]["tn"] += 1
self.total += 1
def recall(self, name):
t = self.h[name]["tp"] + self.h[name]["fn"]
if t == 0:
return 0
return ( (float)(self.h[name]["tp"]) / (float)( t ) )
def precision(self, name):
t = self.h[name]["tp"] + self.h[name]["fp"]
if t == 0:
return 0
return ( (float)(self.h[name]["tp"]) / (float)( t ) )
def fscore(self, name):
t = self.precision(name) + self.recall(name)
if t == 0:
return 0
return (2 * self.precision(name) * self.recall(name) ) / (t )
def score_semeval_2016(self):
return ( self.fscore(0) + self.fscore(1) ) / 2
def score_semeval_2017(self):
return ( self.recall(0) + self.recall(1) + self.recall(2) ) / 3
def score_deft_2017(self, task):
print self.h
if task == "task1":
print "negative : "+str(self.fscore(0))
print "positive : "+str(self.fscore(1))
print "objective : "+str(self.fscore(2))
print "mixed : "+str(self.fscore(3))
return (self.fscore(0) + self.fscore(1) + self.fscore(2) + self.fscore(3) ) / 4
if task == "task3":
print "negative : "+str(self.fscore(0))
print "positive : "+str(self.fscore(1))
print "objective : "+str(self.fscore(2))
print "mixed : "+str(self.fscore(3))
return (self.fscore(0) + self.fscore(1) + self.fscore(2) + self.fscore(3) ) / 4
if task == "task2":
print "figurative : "+str(self.fscore(0))
print "nonfigurative : "+str(self.fscore(1))
return (self.fscore(0) + self.fscore(1) ) / 2
def info(self):
print self.h
def read_word2vec(embedding_file):
ar = []
dico = {}
size = 0
counter = 0
with open(embedding_file) as f:
for line in f:
line = line.strip()
line = line.split(" ")
if len(line) > 3:
ar.append( map(float, line[1:] ) )
dico[ line[0] ] = counter
counter += 1
if len(line) < 3:
size = int(line[1])
ar = np.array(ar, dtype='float')
return counter, size, dico, ar
def read_sentiment_train(sentiment_file, word_word2vec, task):
X_word = []
Y = []
Z = []
Y_COUNTER = None
if task == "task1":
Y_COUNTER = [0.0]*4
if task == "task2":
Y_COUNTER = [0.0]*2
if task == "task3":
Y_COUNTER = [0.0]*4
total = 0
with open(sentiment_file) as f:
for line in f:
line = line.strip()
line = line.split("\t")
if task == "task1":
if line[1] == "negative":
Y_COUNTER[0] += 1
if line[1] == "positive":
Y_COUNTER[1] += 1
if line[1] == "objective":
Y_COUNTER[2] += 1
if line[1] == "mixed":
Y_COUNTER[3] += 1
total += 1
if task == "task3":
if line[1] == "negative":
Y_COUNTER[0] += 1
if line[1] == "positive":
Y_COUNTER[1] += 1
if line[1] == "objective":
Y_COUNTER[2] += 1
if line[1] == "mixed":
Y_COUNTER[3] += 1
total += 1
if task == "task2":
if line[1] == "figurative":
Y_COUNTER[0] += 1
if line[1] == "nonfigurative":
Y_COUNTER[1] += 1
total += 1
print Y_COUNTER
hash_y_counter = None
if task == "task1":
hash_y_counter = {0: Y_COUNTER[0], 1: Y_COUNTER[1], 2: Y_COUNTER[2], 3: Y_COUNTER[3]}
if task == "task3":
hash_y_counter = {0: Y_COUNTER[0], 1: Y_COUNTER[1], 2: Y_COUNTER[2], 3: Y_COUNTER[3]}
if task == "task2":
hash_y_counter = {0: Y_COUNTER[0], 1: Y_COUNTER[1]}
print hash_y_counter
cw_y = create_class_weight(hash_y_counter, 0.5)
print cw_y
with open(sentiment_file) as f:
for line in f:
line = line.strip()
line = line.split("\t")
Z.append( line[0] )
ar_y = None
if task == "task1":
ar_y = [0]*4
if line[1] == "negative":
ar_y[0] = 1
if line[1] == "positive":
ar_y[1] = 1
if line[1] == "objective":
ar_y[2] = 1
if line[1] == "mixed":
ar_y[3] = 1
if task == "task2":
ar_y = [0]*2
if line[1] == "figurative":
ar_y[0] = 1
if line[1] == "nonfigurative":
ar_y[1] = 1
if task == "task3":
ar_y = [0]*4
if line[1] == "negative":
ar_y[0] = 1
if line[1] == "positive":
ar_y[1] = 1
if line[1] == "objective":
ar_y[2] = 1
if line[1] == "mixed":
ar_y[3] = 1
Y.append( ar_y )
tok = line[2].split(" ")
word_ar = []
for x in tok:
if x in word_word2vec:
word_ar.append( word_word2vec[ x ] )
X_word.append( word_ar )
X_word = np.array(X_word)
Y = np.array(Y)
Z = np.array(Z)
return (Z, X_word, Y, cw_y)
def read_sentiment_test(sentiment_file, word_word2vec):
X_word = []
Y = []
Z = []
with open(sentiment_file) as f:
for line in f:
line = line.strip()
line = line.split("\t")
Z.append( line[0] )
ar_y = [0]
Y.append( ar_y )
tok = line[2].split(" ")
word_ar = []
for x in tok:
if x in word_word2vec:
word_ar.append( word_word2vec[ x ] )
X_word.append( word_ar )
X_word = np.array(X_word)
Y = np.array(Y)
Z = np.array(Z)
return (Z, X_word, Y)
def read_sentiment(sentiment_file, word_word2vec, task):
X_word = []
Y = []
Z = []
with open(sentiment_file) as f:
for line in f:
line = line.strip()
line = line.split("\t")
Z.append( line[0] )
ar_y = None
if task == "task1":
ar_y = [0]*4
if line[1] == "negative":
ar_y[0] = 1
if line[1] == "positive":
ar_y[1] = 1
if line[1] == "objective":
ar_y[2] = 1
if line[1] == "mixed":
ar_y[3] = 1
if task == "task2":
ar_y = [0]*2
if line[1] == "figurative":
ar_y[0] = 1
if line[1] == "nonfigurative":
ar_y[1] = 1
if task == "task3":
ar_y = [0]*4
if line[1] == "negative":
ar_y[0] = 1
if line[1] == "positive":
ar_y[1] = 1
if line[1] == "objective":
ar_y[2] = 1
if line[1] == "mixed":
ar_y[3] = 1
Y.append( ar_y )
tok = line[2].split(" ")
word_ar = []
for x in tok:
if x in word_word2vec:
word_ar.append( word_word2vec[ x ] )
X_word.append( word_ar )
X_word = np.array(X_word)
Y = np.array(Y)
Z = np.array(Z)
return (Z, X_word, Y)
maxlen = 150
word_nb_feature_maps = 200
hidden_size = 64
word_embedding_file = sys.argv[1]
test_file = sys.argv[2]
model_file = sys.argv[3]
model = model_from_json(open(model_file+".json").read())
model.load_weights(model_file+".h5")
word_vocab_size, word_embedding_size, word_dico, word_initialize_weight = read_word2vec(word_embedding_file)
ID_train, X_word_train, Y_train = read_sentiment_test(test_file, word_dico)
X_word_train = sequence.pad_sequences(X_word_train, maxlen=maxlen, padding='post', truncating='post')
predict = model.predict(X_word_train, batch_size=2048, verbose=0)
counter = 0
for x in ID_train:
print x+"\t"+" ".join(map(str, predict[counter]))
counter += 1
|