Mercurial > ift6266

comparison deep/stacked_dae/v_sylvain/nist_sda_retrieve.py @ 317:067e747fd9c0

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Ajout de noms differents pour les series produites pour differents choix de pretrain
author SylvainPL <sylvain.pannetier.lebeuf@umontreal.ca>
date Thu, 01 Apr 2010 20:09:14 -0400
parents 60e82846a10d
children 71ffe2c9bfad
comparison
equal deleted inserted replaced
316:60e82846a10d 317:067e747fd9c0
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 #To be sure variables will not be only in the if statement
59 series = create_series(state.num_hidden_layers) 59 PATH = ''
60 nom_reptrain = ''
61 nom_serie = ""
62 if state['pretrain_choice'] == 0:
63 PATH=PATH_NIST
64 nom_pretrain='NIST'
65 nom_serie="series_NIST.h5"
66 elif state['pretrain_choice'] == 1:
67 PATH=PATH_P07
68 nom_pretrain='P07'
69 nom_serie="series_P07.h5"
70
71 series = create_series(state.num_hidden_layers,nom_serie)
60 72
61 print "Creating optimizer with state, ", state 73 print "Creating optimizer with state, ", state
62 74
63 optimizer = SdaSgdOptimizer(dataset=datasets.nist_all(), 75 optimizer = SdaSgdOptimizer(dataset=datasets.nist_all(),
64 hyperparameters=state, \ 76 hyperparameters=state, \
81 ## sys.exit("The code does not support this much pretraining epoch (99 max with P07).\n"+ 93 ## sys.exit("The code does not support this much pretraining epoch (99 max with P07).\n"+
82 ## "You have to correct the code (and be patient, P07 is huge !!)\n"+ 94 ## "You have to correct the code (and be patient, P07 is huge !!)\n"+
83 ## "or reduce the number of pretraining epoch to run the code (better idea).\n") 95 ## "or reduce the number of pretraining epoch to run the code (better idea).\n")
84 ## print('\n\tpretraining with P07') 96 ## print('\n\tpretraining with P07')
85 ## optimizer.pretrain(datasets.nist_P07(min_file=0,max_file=nb_file)) 97 ## optimizer.pretrain(datasets.nist_P07(min_file=0,max_file=nb_file))
86 if state['pretrain_choice'] == 0:
87 PATH=PATH_NIST
88 nom_pretrain='NIST'
89 elif state['pretrain_choice'] == 1:
90 PATH=PATH_P07
91 nom_pretrain='P07'
92 98
93 print ('Retrieve pre-train done earlier ( '+nom_pretrain+' )') 99 print ('Retrieve pre-train done earlier ( '+nom_pretrain+' )')
94 100
95 101
96 102
166 172
167 return channel.COMPLETE 173 return channel.COMPLETE
168 174
169 # These Series objects are used to save various statistics 175 # These Series objects are used to save various statistics
170 # during the training. 176 # during the training.
171 def create_series(num_hidden_layers): 177 def create_series(num_hidden_layers, nom_serie):
172 178
173 # Replace series we don't want to save with DummySeries, e.g. 179 # Replace series we don't want to save with DummySeries, e.g.
174 # series['training_error'] = DummySeries() 180 # series['training_error'] = DummySeries()
175 181
176 series = {} 182 series = {}
177 183
178 basedir = os.getcwd() 184 basedir = os.getcwd()
179 185
180 h5f = tables.openFile(os.path.join(basedir, "series.h5"), "w") 186 h5f = tables.openFile(os.path.join(basedir, nom_serie), "w")
181 187
182 # reconstruction 188 # reconstruction
183 reconstruction_base = \ 189 reconstruction_base = \
184 ErrorSeries(error_name="reconstruction_error", 190 ErrorSeries(error_name="reconstruction_error",
185 table_name="reconstruction_error", 191 table_name="reconstruction_error",