Mercurial > ift6266

comparison deep/convolutional_dae/sgd_opt.py @ 288:80ee63c3e749

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Add net saving (only the best model) and error saving using SeriesTable
author Arnaud Bergeron <abergeron@gmail.com>
date Fri, 26 Mar 2010 17:24:17 -0400
parents 727ed56fad12
children 8babd43235dd
comparison
equal deleted inserted replaced
286:1cc535f3e254 288:80ee63c3e749
1 import time 1 import time
2 import sys 2 import sys
3 3
4 from ift6266.utils.seriestables import *
5
6 default_series = {
7 'train_error' : DummySeries(),
8 'valid_error' : DummySeries(),
9 'test_error' : DummySeries()
10 }
11
4 def sgd_opt(train, valid, test, training_epochs=10000, patience=10000, 12 def sgd_opt(train, valid, test, training_epochs=10000, patience=10000,
5 patience_increase=2., improvement_threshold=0.995, 13 patience_increase=2., improvement_threshold=0.995, net=None,
6 validation_frequency=None): 14 validation_frequency=None, series=default_series):
7 15
8 if validation_frequency is None: 16 if validation_frequency is None:
9 validation_frequency = patience/2 17 validation_frequency = patience/2
10 18
11 start_time = time.clock() 19 start_time = time.clock()
15 test_score = 0. 23 test_score = 0.
16 24
17 start_time = time.clock() 25 start_time = time.clock()
18 26
19 for epoch in xrange(1, training_epochs+1): 27 for epoch in xrange(1, training_epochs+1):
20 train() 28 series['train_error'].append((epoch,), train())
21 29
22 if epoch % validation_frequency == 0: 30 if epoch % validation_frequency == 0:
23 this_validation_loss = valid() 31 this_validation_loss = valid()
32 series['valid_error'].append((epoch,), this_validation_loss*100.)
24 print('epoch %i, validation error %f %%' % \ 33 print('epoch %i, validation error %f %%' % \
25 (epoch, this_validation_loss*100.)) 34 (epoch, this_validation_loss*100.))
26 35
27 # if we got the best validation score until now 36 # if we got the best validation score until now
28 if this_validation_loss < best_validation_loss: 37 if this_validation_loss < best_validation_loss:
36 best_validation_loss = this_validation_loss 45 best_validation_loss = this_validation_loss
37 best_epoch = epoch 46 best_epoch = epoch
38 47
39 # test it on the test set 48 # test it on the test set
40 test_score = test() 49 test_score = test()
50 series['test_error'].append((epoch,), test_score*100.)
41 print((' epoch %i, test error of best model %f %%') % 51 print((' epoch %i, test error of best model %f %%') %
42 (epoch, test_score*100.)) 52 (epoch, test_score*100.))
53 if net is not None:
54 net.save('best.net.new')
55 os.rename('best.net.new', 'best.net')
43 56
44 if patience <= epoch: 57 if patience <= epoch:
45 break 58 break
46 59
47 end_time = time.clock() 60 end_time = time.clock()