Mercurial > ift6266

comparison deep/stacked_dae/v_sylvain/nist_sda.py @ 318:8de3bef71458

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Ajoute plus de fonctionnalite
author SylvainPL <sylvain.pannetier.lebeuf@umontreal.ca>
date Fri, 02 Apr 2010 09:12:40 -0400
parents bd6085d77706
children 7a12d2c3d06b
comparison
equal deleted inserted replaced
317:067e747fd9c0 318:8de3bef71458
53 53
54 n_ins = 32*32 54 n_ins = 32*32
55 n_outs = 62 # 10 digits, 26*2 (lower, capitals) 55 n_outs = 62 # 10 digits, 26*2 (lower, capitals)
56 56
57 examples_per_epoch = NIST_ALL_TRAIN_SIZE 57 examples_per_epoch = NIST_ALL_TRAIN_SIZE
58 58
59 series = create_series(state.num_hidden_layers) 59 #To be sure variables will not be only in the if statement
60 PATH = ''
61 nom_reptrain = ''
62 nom_serie = ""
63 if state['pretrain_choice'] == 0:
64 nom_serie="series_NIST.h5"
65 elif state['pretrain_choice'] == 1:
66 nom_serie="series_P07.h5"
67
68 series = create_series(state.num_hidden_layers,nom_serie)
69
60 70
61 print "Creating optimizer with state, ", state 71 print "Creating optimizer with state, ", state
62 72
63 optimizer = SdaSgdOptimizer(dataset=datasets.nist_all(), 73 optimizer = SdaSgdOptimizer(dataset=datasets.nist_all(),
64 hyperparameters=state, \ 74 hyperparameters=state, \
125 optimizer.reload_parameters('params_pretrain.txt') 135 optimizer.reload_parameters('params_pretrain.txt')
126 optimizer.finetune(datasets.nist_all(),datasets.nist_P07(),max_finetune_epoch_NIST,ind_test=1,special=1) 136 optimizer.finetune(datasets.nist_all(),datasets.nist_P07(),max_finetune_epoch_NIST,ind_test=1,special=1)
127 137
128 138
129 if finetune_choice==-1: 139 if finetune_choice==-1:
130 print('\nSERIE OF 3 DIFFERENT FINETUNINGS') 140 print('\nSERIE OF 4 DIFFERENT FINETUNINGS')
131 print('\n\n\tfinetune with NIST\n\n') 141 print('\n\n\tfinetune with NIST\n\n')
142 sys.stdout.flush()
132 optimizer.reload_parameters('params_pretrain.txt') 143 optimizer.reload_parameters('params_pretrain.txt')
133 optimizer.finetune(datasets.nist_all(),datasets.nist_P07(),max_finetune_epoch_NIST,ind_test=1) 144 optimizer.finetune(datasets.nist_all(),datasets.nist_P07(),max_finetune_epoch_NIST,ind_test=1)
134 channel.save() 145 channel.save()
135 print('\n\n\tfinetune with P07\n\n') 146 print('\n\n\tfinetune with P07\n\n')
147 sys.stdout.flush()
136 optimizer.reload_parameters('params_pretrain.txt') 148 optimizer.reload_parameters('params_pretrain.txt')
137 optimizer.finetune(datasets.nist_P07(),datasets.nist_all(),max_finetune_epoch_P07,ind_test=0) 149 optimizer.finetune(datasets.nist_P07(),datasets.nist_all(),max_finetune_epoch_P07,ind_test=0)
138 channel.save() 150 channel.save()
139 print('\n\n\tfinetune with NIST (done earlier) followed by P07 (written here)\n\n') 151 print('\n\n\tfinetune with NIST (done earlier) followed by P07 (written here)\n\n')
152 sys.stdout.flush()
140 optimizer.reload_parameters('params_finetune_NIST.txt') 153 optimizer.reload_parameters('params_finetune_NIST.txt')
141 optimizer.finetune(datasets.nist_P07(),datasets.nist_all(),max_finetune_epoch_P07,ind_test=20) 154 optimizer.finetune(datasets.nist_P07(),datasets.nist_all(),max_finetune_epoch_P07,ind_test=20)
142 channel.save() 155 channel.save()
143 print('\n\n\tfinetune with NIST only on the logistic regression on top.\n\ 156 print('\n\n\tfinetune with NIST only on the logistic regression on top.\n\
144 All hidden units output are input of the logistic regression\n\n') 157 All hidden units output are input of the logistic regression\n\n')
158 sys.stdout.flush()
145 optimizer.reload_parameters('params_pretrain.txt') 159 optimizer.reload_parameters('params_pretrain.txt')
146 optimizer.finetune(datasets.nist_all(),datasets.nist_P07(),max_finetune_epoch_NIST,ind_test=1,special=1) 160 optimizer.finetune(datasets.nist_all(),datasets.nist_P07(),max_finetune_epoch_NIST,ind_test=1,special=1)
147 channel.save() 161 channel.save()
148 162
149 channel.save() 163 channel.save()
150 164
151 return channel.COMPLETE 165 return channel.COMPLETE
152 166
153 # These Series objects are used to save various statistics 167 # These Series objects are used to save various statistics
154 # during the training. 168 # during the training.
155 def create_series(num_hidden_layers): 169 def create_series(num_hidden_layers, nom_serie):
156 170
157 # Replace series we don't want to save with DummySeries, e.g. 171 # Replace series we don't want to save with DummySeries, e.g.
158 # series['training_error'] = DummySeries() 172 # series['training_error'] = DummySeries()
159 173
160 series = {} 174 series = {}
161 175
162 basedir = os.getcwd() 176 basedir = os.getcwd()
163 177
164 h5f = tables.openFile(os.path.join(basedir, "series.h5"), "w") 178 h5f = tables.openFile(os.path.join(basedir, nom_serie), "w")
165 179
166 # reconstruction 180 # reconstruction
167 reconstruction_base = \ 181 reconstruction_base = \
168 ErrorSeries(error_name="reconstruction_error", 182 ErrorSeries(error_name="reconstruction_error",
169 table_name="reconstruction_error", 183 table_name="reconstruction_error",