Blame view

DECODA_binary_BOW_AEINIT_TANH_MODELS.py 10.2 KB
b6d0165d1   Killian   Initial commit
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
  
  # coding: utf-8
  
  # In[2]:
  
  # Import
  import pandas
  # Alignement
  from alignment.sequence import Sequence
  from alignment.vocabulary import Vocabulary
  from alignment.sequencealigner import *
  import nltk
  import codecs
  import gensim
  from scipy import sparse
  import itertools
  from sklearn.feature_extraction.text import CountVectorizer
  import scipy.sparse
  import scipy.io
  from sklearn import preprocessing
  from keras.models import Sequential
  from keras.layers.core import Dense, Dropout, Activation,AutoEncoder
  from keras.optimizers import SGD,Adam
  from keras.layers import containers
  from mlp import *
  import mlp
  import sklearn.metrics
  import shelve
  import pickle
  from utils import *
  import sys
  # In[4]:
  
  db=shelve.open("{}.shelve".format(sys.argv[1]),writeback=True)
  
  sparse_corp=shelve.open("DECODA_sparse.shelve")
  # In[6]:
  ASR_sparse=sparse_corp["ASR_SPARSE"]
  TRS_sparse=sparse_corp["TRS_SPARSE"]
  
  # In[11]:
  hidden_size=3096
  hidden_size2=2048
  input_activation="tanh"
  out_activation="tanh"
  loss="mse"
  epochs=500
  batch=128
  patience=40
  
  print "gogo autoencoder ASR"
  sgd = SGD(lr=0.0001)#, momentum=0.9, nesterov=True)
  autoencode=Sequential()
  autoencode.add(Dense(ASR_sparse["TRAIN"].shape[1],hidden_size,init='glorot_uniform',activation=input_activation))
  autoencode.add(Dense(hidden_size,hidden_size2,init='glorot_uniform',activation=input_activation))
  autoencode.add(Dense(hidden_size2,hidden_size,init="glorot_uniform",activation=out_activation))
  autoencode.add(Dense(hidden_size,ASR_sparse["DEV"].todense().shape[1],init="glorot_uniform",activation=out_activation))
  
  #autoencode.compile(optimizer=sgd,loss=loss)
  
  autoencode.compile(optimizer=sgd,loss=loss)
  
  
  # In[ ]:
  
  autoencode.fit(ASR_sparse["TRAIN"].todense(),ASR_sparse["TRAIN"].todense(),nb_epoch=epochs,batch_size=batch,
                 callbacks=[keras.callbacks.EarlyStopping(monitor='val_loss',
                                                          patience=patience, verbose=0)],           validation_data=(ASR_sparse["DEV"].todense(),ASR_sparse["DEV"].todense()),verbose=1)
  
  
  # In[ ]:
  
  auto_decoder=Sequential()
  auto_decoder.add(Dense(ASR_sparse["DEV"].todense().shape[1],hidden_size,init='uniform',activation=input_activation,weights=autoencode.get_weights()[:2]))
  auto_decoder.add(Dense(hidden_size,hidden_size2,init='glorot_uniform',activation=input_activation,weights=autoencode.get_weights()[2:4]))
  auto_decoder.add(Dense(hidden_size2,hidden_size,init='glorot_uniform',activation=input_activation,weights=autoencode.get_weights()[4:6]))
  auto_decoder.compile(optimizer=sgd,loss=loss)
  
  
  # In[77]:
  
  #autoencode.predict(ASR_sparse["DEV"].todense())
  
  
  # In[ ]:
  
  print "auto encoder et auto decoder asr okay"
  
  ASR_sparse_AE={}
  for i in ASR_sparse.keys():
      ASR_sparse_AE[i]=auto_decoder.predict(ASR_sparse[i].todense())
      #TRS_sparse[i]=dico.transform(TRS[i][2])
  
  db["ASR_AE_H2_RELU"]=ASR_sparse_AE
  
  
  auto_decoder=Sequential()
  auto_decoder.add(Dense(ASR_sparse["DEV"].todense().shape[1],hidden_size,init='uniform',activation=input_activation,weights=autoencode.get_weights()[:2]))
  auto_decoder.add(Dense(hidden_size,hidden_size2,init='glorot_uniform',activation=input_activation,weights=autoencode.get_weights()[2:4]))
  auto_decoder.compile(optimizer=sgd,loss=loss)
  
  ASR_sparse_AE_H1={}
  for i in ASR_sparse.keys():
      ASR_sparse_AE_H1[i]=auto_decoder.predict(ASR_sparse[i].todense())
      #TRS_sparse[i]=dico.transform(TRS[i][2])
  
  db["ASR_AE_H1_RELU"]=ASR_sparse_AE_H1
  
  
  
  print "auto encoder trs learning"
  # In[68]:/
  sgd_trs = SGD(lr=0.1,momentum=0.9)
  autoencode_trs=Sequential()
  autoencode_trs.add(Dense(TRS_sparse["DEV"].todense().shape[1],hidden_size,init='glorot_uniform',activation=input_activation))
  autoencode_trs.add(Dense(hidden_size,hidden_size2,init='glorot_uniform',activation=input_activation))
  autoencode_trs.add(Dense(hidden_size2,hidden_size,init="glorot_uniform",activation=out_activation))
  autoencode_trs.add(Dense(hidden_size,TRS_sparse["DEV"].todense().shape[1],init="glorot_uniform",activation=out_activation))
  
  #autoencode_trs.compile(optimizer=sgd_trs,loss=loss)
  
  autoencode_trs.compile(optimizer=sgd,loss=loss)
  
  
  # In[69]:
  
  autoencode_trs.fit(TRS_sparse["TRAIN"].todense(),TRS_sparse["TRAIN"].todense(),nb_epoch=epochs,batch_size=batch,
                 callbacks=[keras.callbacks.EarlyStopping(monitor='val_loss',
                                                          patience=patience, verbose=0)],
                 validation_data=(TRS_sparse["DEV"].todense(),TRS_sparse["DEV"].todense()),verbose=1)
  
  
  # In[87]:
  
  auto_decoder_trs=Sequential()
  auto_decoder_trs.add(Dense(ASR_sparse["DEV"].todense().shape[1],hidden_size,activation=input_activation,weights=autoencode_trs.get_weights()[:2]))
  auto_decoder_trs.add(Dense(hidden_size,hidden_size2,activation=input_activation,weights=autoencode_trs.get_weights()[2:4]))
  auto_decoder_trs.add(Dense(hidden_size2,hidden_size,activation=input_activation,weights=autoencode_trs.get_weights()[4:6]))
  auto_decoder_trs.compile(optimizer=sgd,loss=loss)
  
  
  # In[88]:
  print "auto encoder trs okay"
  TRS_sparse_AE={}
  
  for i in TRS_sparse.keys():
      TRS_sparse_AE[i]=auto_decoder_trs.predict(TRS_sparse[i].todense())
      #TRS_sparse[i]=dico.transform(TRS[i][2])
  
  db["TRS_AE_H2_RELU"]=TRS_sparse_AE
  
  
  
  auto_decoder_trs=Sequential()
  auto_decoder_trs.add(Dense(ASR_sparse["DEV"].todense().shape[1],hidden_size,activation=input_activation,weights=autoencode_trs.get_weights()[:2]))
  auto_decoder_trs.add(Dense(hidden_size,hidden_size2,activation=input_activation,weights=autoencode_trs.get_weights()[2:4]))
  auto_decoder_trs.compile(optimizer=sgd,loss=loss)
  
  
  # In[88]:
  print "auto encoder trs okay"
  TRS_sparse_AE_H1={}
  
  for i in TRS_sparse.keys():
      TRS_sparse_AE_H1[i]=auto_decoder_trs.predict(TRS_sparse[i].todense())
      #TRS_sparse[i]=dico.transform(TRS[i][2])
  
  db["TRS_AE_H1_RELU"]=TRS_sparse_AE
  
  
  db.sync()
  
  
  
  
  # In[261]:
  
  #pred_dev= model_TRS_AE.predict(TRS_sparse_AE["DEV"],batch_size=1)
  
  TRS_AE={}
  ASR_AE={}
  for i in TRS_sparse.keys():
      TRS_AE[i]=autoencode_trs.predict(TRS_sparse[i].todense())
      ASR_AE[i]=autoencode.predict(ASR_sparse[i].todense())
  
  
  db["TRS_AE_OUT_RELU"]=TRS_AE
  db["ASR_AE_OUT_RELU"]=ASR_AE
  
  db.sync()
  # # Transfert de couche
  # ICI
  spe_finetune=Sequential()
  spe_finetune.add(Dense(TRS_sparse["DEV"].todense().shape[1],hidden_size,init='glorot_uniform',activation=input_activation,weights=autoencode.get_weights()[:2]))
  spe_finetune.add(Dense(hidden_size,hidden_size2,init='glorot_uniform',activation=input_activation,weights=autoencode.get_weights()[2:4]))
  spe_finetune.add(Dense(hidden_size2,hidden_size,init="glorot_uniform",activation=out_activation,weights=autoencode_trs.get_weights()[4:6]))
  spe_finetune.add(Dense(hidden_size,TRS_sparse["DEV"].todense().shape[1],init="glorot_uniform",activation=out_activation,weights=autoencode_trs.get_weights()[6:8]))
  
  #spe_finetune.compile(optimizer=sgd_trs,loss=loss)
  
  spe_finetune.compile(optimizer=sgd,loss=loss)
  
  
  auto_decoder_spe_finetune=Sequential()
  auto_decoder_spe_finetune.add(Dense(ASR_sparse["DEV"].todense().shape[1],hidden_size,activation=input_activation,weights=autoencode.get_weights()[:2]))
  auto_decoder_spe_finetune.add(Dense(hidden_size,hidden_size2,activation=input_activation,weights=autoencode.get_weights()[2:4]))
  auto_decoder_spe_finetune.add(Dense(hidden_size2,hidden_size,init="glorot_uniform",activation=out_activation,weights=autoencode_trs.get_weights()[4:6]))
  auto_decoder_spe_finetune.compile(optimizer=sgd,loss=loss)
  
  
  
  spe_finetune_fixed=Sequential()
  spe_finetune_fixed.add(Dense(hidden_size2,hidden_size,init="glorot_uniform",activation=out_activation,weights=autoencode_trs.get_weights()[4:6]))
  spe_finetune_fixed.add(Dense(hidden_size,TRS_sparse["DEV"].todense().shape[1],init="glorot_uniform",activation=out_activation,weights=autoencode_trs.get_weights()[6:8]))
  spe_finetune_fixed.compile(optimizer=sgd,loss=loss)
  
  
  # In[88]:
  print "auto encoder SPE RAW OKAY"
  ASR_AE_SPE_FINETUNE_RAW_OUT={}
  ASR_AE_SPE_FINETUNE_RAW_H2={}
  ASR_AE_SPE_FINETUNE_RAW_FIXED_OUT={}
  for i in TRS_sparse.keys():
      ASR_AE_SPE_FINETUNE_RAW_OUT[i]=spe_finetune.predict(ASR_sparse[i].todense())
      #TRS_sparse[i]=dico.transform(TRS[i][2])
      ASR_AE_SPE_FINETUNE_RAW_H2[i]=auto_decoder_spe_finetune.predict(ASR_sparse[i].todense())
      ASR_AE_SPE_FINETUNE_RAW_FIXED_OUT[i]=spe_finetune_fixed.predict(ASR_sparse_AE_H1[i])
  
  db["ASR_AE_SPE_FINETUNE_RAW_OUT"]=ASR_AE_SPE_FINETUNE_RAW_OUT
  db["ASR_AE_SPE_FINETUNE_RAW_H2"]=ASR_AE_SPE_FINETUNE_RAW_H2
  db["ASR_AE_SPE_FINETUNE_RAW_FIXED_OUT"]=ASR_AE_SPE_FINETUNE_RAW_FIXED_OUT
  db.sync()
  
  
  spe_finetune.fit(ASR_sparse["TRAIN"].todense(),TRS_sparse["TRAIN"].todense(),nb_epoch=epochs,batch_size=batch,
                 callbacks=[keras.callbacks.EarlyStopping(monitor='val_loss',
                                                          patience=patience, verbose=0)],           validation_data=(ASR_sparse["DEV"].todense(),TRS_sparse["DEV"].todense()),verbose=1)
  
  
  
  auto_decoder_spe_finetune=Sequential()
  auto_decoder_spe_finetune.add(Dense(ASR_sparse["DEV"].todense().shape[1],hidden_size,activation=input_activation,weights=spe_finetune.get_weights()[:2]))
  auto_decoder_spe_finetune.add(Dense(hidden_size,hidden_size2,activation=input_activation,weights=spe_finetune.get_weights()[2:4]))
  auto_decoder_spe_finetune.add(Dense(hidden_size2,hidden_size,init="glorot_uniform",activation=out_activation,weights=spe_finetune.get_weights()[4:6]))
  auto_decoder_spe_finetune.compile(optimizer=sgd,loss=loss)
  
  
  ASR_sparse_AE_H1[i]
  
  
  spe_finetune_fixed.fit(ASR_sparse_AE_H1["TRAIN"],TRS_sparse["TRAIN"].todense(),nb_epoch=epochs,batch_size=batch,callbacks=[keras.callbacks.EarlyStopping(monitor='val_loss',
      patience=patience, verbose=0)],
      validation_data=(ASR_sparse_AE_H1["DEV"],TRS_sparse["DEV"].todense()),verbose=1)
  
  
  ASR_AE_SPE_FINETUNE_TUNED_OUT={}
  ASR_AE_SPE_FINETUNE_TUNED_H2={}
  ASR_AE_SPE_FINETUNE_TUNED_FIXED_OUT={}
  
  for i in TRS_sparse.keys():
      ASR_AE_SPE_FINETUNE_TUNED_OUT[i]=spe_finetune.predict(ASR_sparse[i].todense())
      #TRS_sparse[i]=dico.transform(TRS[i][2])
      ASR_AE_SPE_FINETUNE_TUNED_H2[i]=auto_decoder_spe_finetune.predict(ASR_sparse[i].todense())
      ASR_AE_SPE_FINETUNE_TUNED_FIXED_OUT[i]=spe_finetune_fixed.predict(ASR_sparse_AE_H1[i])
  db["ASR_AE_SPE_FINETUNE_TUNED_OUT"]=ASR_AE_SPE_FINETUNE_RAW_OUT
  db["ASR_AE_SPE_FINETUNE_TUNED_H2"]=ASR_AE_SPE_FINETUNE_RAW_H2
  db["ASR_AE_SPE_FINETUNE_TUNED_FIXED_OUT"]=ASR_AE_SPE_FINETUNE_RAW_FIXED_OUT
  
  db.sync()
  db.close()