Mercurial > ift6266

comparison deep/stacked_dae/nist_sda.py @ 284:8a3af19ae272

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Enlevé mécanique pour limiter le nombre d'exemples utilisés (remplacé par paramètre dans l'appel au code de dataset), et ajouté option pour sauvegarde des poids à la fin de l'entraînement
author fsavard
date Wed, 24 Mar 2010 15:13:48 -0400
parents 206374eed2fb
children
comparison
equal deleted inserted replaced
279:206374eed2fb 284:8a3af19ae272
54 54
55 n_ins = 32*32 55 n_ins = 32*32
56 n_outs = 62 # 10 digits, 26*2 (lower, capitals) 56 n_outs = 62 # 10 digits, 26*2 (lower, capitals)
57 57
58 examples_per_epoch = NIST_ALL_TRAIN_SIZE 58 examples_per_epoch = NIST_ALL_TRAIN_SIZE
59 if rtt:
60 examples_per_epoch = rtt
59 61
60 series = create_series(state.num_hidden_layers) 62 series = create_series(state.num_hidden_layers)
61 63
62 print "Creating optimizer with state, ", state 64 print "Creating optimizer with state, ", state
63 65
64 optimizer = SdaSgdOptimizer(dataset=datasets.nist_all(), 66 dataset = None
67 if rtt:
68 dataset = datasets.nist_all(maxsize=rtt)
69 else:
70 dataset = datasets.nist_all()
71
72 optimizer = SdaSgdOptimizer(dataset=dataset,
65 hyperparameters=state, \ 73 hyperparameters=state, \
66 n_ins=n_ins, n_outs=n_outs,\ 74 n_ins=n_ins, n_outs=n_outs,\
67 examples_per_epoch=examples_per_epoch, \ 75 examples_per_epoch=examples_per_epoch, \
68 series=series, 76 series=series,
69 max_minibatches=rtt) 77 save_params=SAVE_PARAMS)
70 78
71 optimizer.pretrain(datasets.nist_all()) 79 optimizer.pretrain(dataset)
72 channel.save() 80 channel.save()
73 81
74 optimizer.finetune(datasets.nist_all()) 82 optimizer.finetune(dataset)
75 channel.save() 83 channel.save()
76 84
77 return channel.COMPLETE 85 return channel.COMPLETE
78 86
79 # These Series objects are used to save various statistics 87 # These Series objects are used to save various statistics