-
Notifications
You must be signed in to change notification settings - Fork 3
/
nn_lm_new.py
307 lines (239 loc) · 13.6 KB
/
nn_lm_new.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
'''
Created on 21 Feb 2013
@author: s1264845
'''
import theano
import theano.tensor as T
import time, numpy
from vocab_create import Vocabulary
from data_provider import DataProvider
import os, sys
from learn_rates import LearningRateNewBob, LearningRateList
from learn_rates import LearningRate
from logistic_regression import LogisticRegression
from mlp import MLP
from mlp import HiddenLayer
from cache import TNetsCacheSimple, TNetsCacheLastElem
from numpy.core.numeric import dtype
from utils import GlobalCfg
from mlp_save import save_mlp, save_posteriors
import math
def train_mlp(L1_reg = 0.0, L2_reg = 0.0000, num_batches_per_bunch = 512, batch_size = 1, num_bunches_queue = 5, offset = 0, path_name = '/afs/inf.ed.ac.uk/user/s12/s1264845/scratch/s1264845/data/'):
voc_list = Vocabulary(path_name + 'train')
voc_list.vocab_create()
vocab = voc_list.vocab
vocab_size = voc_list.vocab_size
voc_list_valid = Vocabulary(path_name + 'valid')
voc_list_valid.vocab_create()
count = voc_list_valid.count
voc_list_test = Vocabulary(path_name + 'test')
voc_list_test.vocab_create()
no_test_tokens = voc_list_test.count
print 'The number of sentenses in test set:', no_test_tokens
#print 'number of words in valid data:', count
dataprovider_train = DataProvider(path_name + 'train', vocab, vocab_size )
dataprovider_valid = DataProvider(path_name + 'valid', vocab, vocab_size )
dataprovider_test = DataProvider(path_name + 'test', vocab, vocab_size )
#learn_list = [0.1, 0.1, 0.1, 0.75, 0.5, 0.25, 0.125, 0.0625, 0]
exp_name = 'fine_tuning.hdf5'
posterior_path = 'log_likelihoods'
print '..building the model'
#symbolic variables for input, target vector and batch index
index = T.lscalar('index')
x = T.fmatrix('x')
y = T.ivector('y')
learning_rate = T.fscalar('learning_rate')
#theano shares variables for train, valid and test
train_set_x = theano.shared(numpy.empty((1,1), dtype='float32'), allow_downcast = True)
train_set_y = theano.shared(numpy.empty((1), dtype = 'int32'), allow_downcast = True)
valid_set_x = theano.shared(numpy.empty((1,1), dtype='float32'), allow_downcast = True)
valid_set_y = theano.shared(numpy.empty((1), dtype = 'int32'), allow_downcast = True)
test_set_x = theano.shared(numpy.empty((1,1), dtype='float32'), allow_downcast = True)
test_set_y = theano.shared(numpy.empty((1), dtype = 'int32'), allow_downcast = True)
rng = numpy.random.RandomState(1234)
classifier = MLP(rng = rng, input = x, n_in = vocab_size, n_hidden1 = 30, n_hidden2 = 60 , n_out = vocab_size)
#classifier = MLP(rng = rng, input = x, n_in = vocab_size, n_hidden = 60, n_out = vocab_size)
cost = classifier.negative_log_likelihood(y) + L1_reg * classifier.L1 + L2_reg * classifier.L2_sqr
#constructor for learning rate class
learnrate_schedular = LearningRateNewBob(start_rate=0.001, scale_by=.5, max_epochs=9999,\
min_derror_ramp_start=.1, min_derror_stop=.1, init_error=100.)
#learnrate_schedular = LearningRateList(learn_list)
frame_error = classifier.errors(y)
likelihood = classifier.sum(y)
#test model
test_model = theano.function(inputs = [index], outputs = likelihood, \
givens = {x: test_set_x[index * batch_size:(index + 1) * batch_size],
y: test_set_y[index * batch_size:(index + 1) * batch_size]})
#validation_model
validate_model = theano.function(inputs = [index], outputs = [frame_error, likelihood], \
givens = {x: valid_set_x[index * batch_size:(index + 1) * batch_size],
y: valid_set_y[index * batch_size:(index + 1) * batch_size]})
gradient_param = []
#calculates the gradient of cost with respect to parameters
for param in classifier.params:
gradient_param.append(T.cast(T.grad(cost, param), 'float32'))
updates = []
for param, gradient in zip(classifier.params, gradient_param):
updates.append((param, param - learning_rate * gradient))
#training_model
train_model = theano.function(inputs = [index, theano.Param(learning_rate, default = 0.01)], outputs = cost, updates = updates, \
givens = {x: train_set_x[index * batch_size:(index + 1) * batch_size],
y: train_set_y[index * batch_size:(index + 1) * batch_size]})
#theano.printing.pydotprint(train_model, outfile = "pics/train.png", var_with_name_simple = True)
#path_save = '/afs/inf.ed.ac.uk/user/s12/s1264845/scratch/s1264845/mlp/saved_weights/'
print '.....training'
best_valid_loss = numpy.inf
epoch = 1
start_time = time.time()
while(learnrate_schedular.get_rate() != 0):
print 'learning_rate:', learnrate_schedular.get_rate()
print 'epoch_number:', learnrate_schedular.epoch
frames_showed, progress = 0, 0
start_epoch_time = time.time()
tqueue = TNetsCacheSimple.make_queue()
cache = TNetsCacheSimple(tqueue, shuffle_frames = True, offset=0, \
batch_size = batch_size, num_batches_per_bunch = num_batches_per_bunch)
cache.data_provider = dataprovider_train
cache.start()
train_cost = []
while True:
feats_lab_tuple = TNetsCacheSimple.get_elem_from_queue(tqueue)
if isinstance(feats_lab_tuple, TNetsCacheLastElem):
break
features, labels = feats_lab_tuple
train_set_x.set_value(features, borrow=True)
train_set_y.set_value(numpy.asarray(labels.flatten(), dtype = 'int32'), borrow=True)
frames_showed += features.shape[0]
train_batches = features.shape[0]/batch_size
#print train_batches
#if there is any part left in utterance (smaller than a batch_size), take it into account at the end
if(features.shape[0] % batch_size!=0 or features.shape[0] < batch_size):
train_batches += 1
for i in xrange(train_batches):
#train_cost.append(train_model(i, learnrate_schedular.get_rate()))
train_model(i, learnrate_schedular.get_rate())
progress += 1
if progress%10==0:
end_time_progress = time.time()
print 'PROGRESS: Processed %i bunches (%i frames), TIME: %f in seconds'\
%(progress, frames_showed,(end_time_progress-start_epoch_time))
end_time_progress = time.time()
print 'PROGRESS: Processed %i bunches (%i frames), TIME: %f in seconds'\
%(progress, frames_showed,(end_time_progress-start_epoch_time))
train_set_x.set_value(numpy.empty((1,1), dtype = 'float32'))
train_set_y.set_value(numpy.empty((1), dtype = 'int32'))
classifier_name = 'MLP' + str(learnrate_schedular.epoch)
save_mlp(classifier, GlobalCfg.get_working_dir()+exp_name , classifier_name)
print 'Validating...'
valid_losses = []
log_likelihood = []
valid_frames_showed, progress = 0, 0
start_valid_time = time.time() # it is also stop of training time
#for feat_lab_tuple, path in HDFDatasetDataProviderUtt(devel_files_list, valid_dataset, randomize=False, max_utt=-10):
# features, labels = feat_lab_tuple
tqueue = TNetsCacheSimple.make_queue()
cache = TNetsCacheSimple(tqueue, offset = 0, num_batches_per_bunch = 16)
#cache.deamon = True
cache.data_provider = dataprovider_valid
cache.start()
#ex_num = 0
while True:
feats_lab_tuple = TNetsCacheSimple.get_elem_from_queue(tqueue)
if isinstance(feats_lab_tuple, TNetsCacheLastElem):
break
features, labels = feats_lab_tuple
valid_frames_showed += features.shape[0]
valid_set_x.set_value(features, borrow=True)
valid_set_y.set_value(numpy.asarray(labels.flatten(), 'int32'), borrow=True)
valid_batches = features.shape[0] / batch_size
#print valid_batches
#if there is any part left in utterance (smaller than a batch_size), take it into account at the end
if(features.shape[0] % batch_size!=0 or features.shape[0] < batch_size):
valid_batches += 1
for i in xrange(valid_batches):
#ex_num = ex_num + 1
out = validate_model(i)
error_rate = out[0]
likelihoods = out[1]
valid_losses.append(error_rate)
log_likelihood.append(likelihoods)
#save_posteriors(likelihoods, GlobalCfg.get_working_dir() + posterior_path, str(ex_num), str(learnrate_schedular.epoch))
progress += 1
if progress%10==0:
end_time_valid_progress = time.time()
print 'PROGRESS: Processed %i bunches (%i frames), TIME: %f in seconds'\
%(progress, valid_frames_showed, end_time_valid_progress - start_valid_time)
end_time_valid_progress = time.time()
print 'PROGRESS: Processed %i bunches (%i frames), TIME: %f in seconds'\
%(progress, valid_frames_showed, end_time_valid_progress - start_valid_time)
valid_set_x.set_value(numpy.empty((1,1), 'float32'))
valid_set_y.set_value(numpy.empty((1), 'int32'))
end_epoch_time = time.time()
print 'time taken for this epoch in seconds: %f' %(end_epoch_time - start_epoch_time)
this_validation_loss = numpy.mean(valid_losses)
loglikelihood_sum = numpy.sum(log_likelihood)
#ppl = math.exp(- loglikelihood_sum /count)
#print 'ppl:', ppl
print 'error_rate:', this_validation_loss
print 'valid log likelihood:', loglikelihood_sum
#print 'mean log_probability', this_validation_loss
#learnrate_schedular.get_next_rate(this_validation_loss * 100.)
#learnrate_schedular.get_next_rate()
#print 'epoch_number:', learnrate_schedular.epoch
# logger.info('Epoch %i (lr: %f) took %f min (SPEED [presentations/second] training %f, cv %f), cv error %f %%' % \
# (self.cfg.finetune_scheduler.epoch-1, self.cfg.finetune_scheduler.get_rate(), \
# ((end_epoch_time-start_epoch_time)/60.0), (frames_showed/(start_valid_time-start_epoch_time)), \
# (valid_frames_showed/(stop_valid_time-start_valid_time)), this_validation_loss*100.))
#self.cfg.finetune_scheduler.get_next_rate(this_validation_loss*100.)
if this_validation_loss < best_valid_loss:
learning_rate = learnrate_schedular.get_next_rate(this_validation_loss * 100.)
best_valid_loss = this_validation_loss
#best_epoch = learnrate_schedular.epoch-1
else:
#learnrate_schedular.epoch = learnrate_schedular.epoch + 1
learnrate_schedular.rate = 0.0
end_time = time.time()
#print 'Optimization complete with best validation score of %f %%' % best_valid_loss * 100.
print 'The fine tuning ran for %.2fm' %((end_time-start_time)/60.)
print 'Testing...'
log_likelihood_test = []
test_frames_showed, progress = 0, 0
start_test_time = time.time() # it is also stop of training time
#for feat_lab_tuple, path in HDFDatasetDataProviderUtt(devel_files_list, valid_dataset, randomize=False, max_utt=-10):
# features, labels = feat_lab_tuple
tqueue = TNetsCacheSimple.make_queue()
cache = TNetsCacheSimple(tqueue, offset = 0, num_batches_per_bunch = 16)
#cache.deamon = True
cache.data_provider = dataprovider_test
cache.start()
#ex_num = 0
while True:
feats_lab_tuple = TNetsCacheSimple.get_elem_from_queue(tqueue)
if isinstance(feats_lab_tuple, TNetsCacheLastElem):
break
features, labels = feats_lab_tuple
test_frames_showed += features.shape[0]
test_set_x.set_value(features, borrow=True)
test_set_y.set_value(numpy.asarray(labels.flatten(), 'int32'), borrow=True)
test_batches = features.shape[0] / batch_size
#print valid_batches
#if there is any part left in utterance (smaller than a batch_size), take it into account at the end
if(features.shape[0] % batch_size!=0 or features.shape[0] < batch_size):
test_batches += 1
for i in xrange(test_batches):
log_likelihood_test.append(test_model(i))
progress += 1
if progress%10==0:
end_time_test_progress = time.time()
print 'PROGRESS: Processed %i bunches (%i frames), TIME: %f in seconds'\
%(progress, test_frames_showed, end_time_test_progress - start_test_time)
end_time_test_progress = time.time()
print 'PROGRESS: Processed %i bunches (%i frames), TIME: %f in seconds'\
%(progress, test_frames_showed, end_time_test_progress - start_test_time)
test_set_x.set_value(numpy.empty((1,1), 'float32'))
test_set_y.set_value(numpy.empty((1), 'int32'))
likelihood_sum = numpy.sum(log_likelihood_test)
print 'likelihood_sum', likelihood_sum
#test_ppl = math.exp(- likelihood_sum / no_test_tokens)
#print 'test_ppl:', test_ppl
train_mlp()