Mercurial > ift6266

comparison deep/stacked_dae/v_sylvain/sgd_optimization.py @ 286:1cc535f3e254

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
correction d'un bug pour affichage des resultats de pre-train avec P07
author SylvainPL <sylvain.pannetier.lebeuf@umontreal.ca>
date Thu, 25 Mar 2010 12:20:27 -0400
parents 28b628f331b2
children fe5d428c2acc
comparison
equal deleted inserted replaced
285:694e75413413 286:1cc535f3e254
85 self.finetune(self.dataset) 85 self.finetune(self.dataset)
86 86
87 def pretrain(self,dataset): 87 def pretrain(self,dataset):
88 print "STARTING PRETRAINING, time = ", datetime.datetime.now() 88 print "STARTING PRETRAINING, time = ", datetime.datetime.now()
89 sys.stdout.flush() 89 sys.stdout.flush()
90
91 un_fichier=int(819200.0/self.hp.minibatch_size) #Number of batches in a P07 batch
90 92
91 start_time = time.clock() 93 start_time = time.clock()
92 ## Pre-train layer-wise 94 ## Pre-train layer-wise
93 for i in xrange(self.classifier.n_layers): 95 for i in xrange(self.classifier.n_layers):
94 # go through pretraining epochs 96 # go through pretraining epochs
111 if self.max_minibatches and batch_index >= self.max_minibatches: 113 if self.max_minibatches and batch_index >= self.max_minibatches:
112 break 114 break
113 115
114 #When we pass through the data only once (the case with P07) 116 #When we pass through the data only once (the case with P07)
115 #There is approximately 800*1024=819200 examples per file (1k per example and files are 800M) 117 #There is approximately 800*1024=819200 examples per file (1k per example and files are 800M)
116 if self.hp.pretraining_epochs_per_layer == 1 and count%819200 == 0: 118 if self.hp.pretraining_epochs_per_layer == 1 and count%un_fichier == 0:
117 print 'Pre-training layer %i, epoch %d, cost '%(i,num_files),c 119 print 'Pre-training layer %i, epoch %d, cost '%(i,num_files),c
118 num_files+=1 120 num_files+=1
119 sys.stdout.flush() 121 sys.stdout.flush()
120 self.series['params'].append((num_files,), self.classifier.all_params) 122 self.series['params'].append((num_files,), self.classifier.all_params)
121 123