Mercurial > pylearn

comparison algorithms/tests/test_daa.py @ 501:4fb6f7320518

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
N-class logistic regression top-layer works
author Joseph Turian <turian@gmail.com>
date Tue, 28 Oct 2008 13:54:01 -0400
parents 3c60c2db0319
children de974b4fc4ea
comparison
equal deleted inserted replaced
500:3c60c2db0319 501:4fb6f7320518
21 model.layers[0].noise_level = 0.3 21 model.layers[0].noise_level = 0.3
22 model.layers[1].noise_level = 0.3 22 model.layers[1].noise_level = 0.3
23 model.layers[2].noise_level = 0.3 23 model.layers[2].noise_level = 0.3
24 24
25 # Update the first hidden layer 25 # Update the first hidden layer
26 model.local_update[0]([[0, 1, 0, 1]]) 26 for l in range(3):
27 model.local_update[1]([[0, 1, 0, 1]]) 27 for i in range(10):
28 model.local_update[2]([[0, 1, 0, 1]]) 28 model.local_update[l]([[0, 1, 0, 1]])
29 model.local_update[l]([[1, 0, 1, 0]])
29 30
30 model.update([[0, 1, 0, 1]], [[0]]) 31 for i in range(1):
32 model.update([[0, 1, 0, 1]], [[1]])
33 model.update([[1, 0, 1, 0]], [[0]])
31 print model.classify([[0, 1, 0, 1]]) 34 print model.classify([[0, 1, 0, 1]])
35 print model.classify([[1, 0, 1, 0]])
32 36
33 37
34 def test_train_daa2(mode = theano.Mode('c|py', 'fast_run')): 38 def test_train_daa2(mode = theano.Mode('c|py', 'fast_run')):
35 39
36 ndaa = 3 40 ndaa = 3
37 daa = models.Stacker([(models.SigmoidXEDenoisingAA, 'hidden')] * ndaa + [(pylearn.algorithms.logistic_regression.Module_Nclass, 'output')], 41 daa = models.Stacker([(models.SigmoidXEDenoisingAA, 'hidden')] * ndaa + [(pylearn.algorithms.logistic_regression.Module_Nclass, 'pred')],
38 regularize = False) 42 regularize = False)
39 43
40 model = daa.make([4, 20, 20, 20, 1], 44 model = daa.make([4, 20, 20, 20, 10],
41 lr = 0.01, 45 lr = 0.01,
42 mode = mode, 46 mode = mode,
43 seed = 10) 47 seed = 10)
44 48
45 model.layers[0].noise_level = 0.3 49 model.layers[0].noise_level = 0.3
46 model.layers[1].noise_level = 0.3 50 model.layers[1].noise_level = 0.3
47 model.layers[2].noise_level = 0.3 51 model.layers[2].noise_level = 0.3
48 52
49 # Update the first hidden layer 53 for l in range(3):
50 model.local_update[0]([[0, 1, 0, 1]]) 54 for i in range(10):
51 model.local_update[1]([[0, 1, 0, 1]]) 55 model.local_update[l]([[0, 1, 0, 1]])
52 model.local_update[2]([[0, 1, 0, 1]]) 56 model.local_update[l]([[1, 0, 1, 0]])
53 57
54 model.update([[0, 1, 0, 1]], [0]) 58 for i in range(1):
55 print model.classify([[0, 1, 0, 1]]) 59 model.update([[0, 1, 0, 1]], [1])
56 60 model.update([[1, 0, 1, 0]], [0])
61 print model.apply([[0, 1, 0, 1]])
62 print model.apply([[1, 0, 1, 0]])
57 63
58 64
59 65
60 66
61 if __name__ == '__main__': 67 if __name__ == '__main__':
68 # print 'not optimized:' 74 # print 'not optimized:'
69 # t2 = test_train_daa(theano.Mode('c|py', 'fast_compile')) 75 # t2 = test_train_daa(theano.Mode('c|py', 'fast_compile'))
70 ## print 'time:',t2 76 ## print 'time:',t2
71 77
72 # test_train_daa(theano.compile.Mode('c&py', 'merge')) 78 # test_train_daa(theano.compile.Mode('c&py', 'merge'))
73 test_train_daa(theano.compile.Mode('c|py', 'merge')) 79 # test_train_daa(theano.compile.Mode('c|py', 'merge'))
80 test_train_daa(theano.compile.Mode('py', 'merge'))
81
74 test_train_daa2(theano.compile.Mode('c|py', 'merge')) 82 test_train_daa2(theano.compile.Mode('c|py', 'merge'))