Mercurial > ift6266

comparison deep/autoencoder/DA_training.py @ 206:e12702b88a2d

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
removed one bug: the complexity value was not sent to the dA class
author youssouf
date Thu, 04 Mar 2010 09:43:23 -0500
parents 70a9df1cd20e
children
comparison
equal deleted inserted replaced
190:70a9df1cd20e 206:e12702b88a2d
91 91
92 # create a Theano random generator that gives symbolic random values 92 # create a Theano random generator that gives symbolic random values
93 theano_rng = RandomStreams() 93 theano_rng = RandomStreams()
94 # create a numpy random generator 94 # create a numpy random generator
95 numpy_rng = numpy.random.RandomState() 95 numpy_rng = numpy.random.RandomState()
96 96
97 # print the parameter of the DA
98 if True :
99 print 'input size = %d' %n_visible
100 print 'hidden size = %d' %n_hidden
101 print 'complexity = %2.2f' %complexity
97 102
98 # initial values for weights and biases 103 # initial values for weights and biases
99 # note : W' was written as `W_prime` and b' as `b_prime` 104 # note : W' was written as `W_prime` and b' as `b_prime`
100 105
101 # W is initialized with `initial_W` which is uniformely sampled 106 # W is initialized with `initial_W` which is uniformely sampled
248 y = T.lvector() # the labels are presented as 1D vector of 253 y = T.lvector() # the labels are presented as 1D vector of
249 # [long int] labels 254 # [long int] labels
250 255
251 # construct the denoising autoencoder class 256 # construct the denoising autoencoder class
252 n_ins = 32*32 257 n_ins = 32*32
253 encoder = dA(n_ins, n_code_layer, input = x.reshape((batch_size,n_ins))) 258 encoder = dA(n_ins, n_code_layer, complexity, input = x.reshape((batch_size,n_ins)))
254 259
255 # Train autoencoder 260 # Train autoencoder
256 261
257 # compute gradients of the layer parameters 262 # compute gradients of the layer parameters
258 gW = T.grad(encoder.cost, encoder.W) 263 gW = T.grad(encoder.cost, encoder.W)
361 'model %f ') % 366 'model %f ') %
362 (epoch, minibatch_index+1, n_minibatches, 367 (epoch, minibatch_index+1, n_minibatches,
363 test_score)) 368 test_score))
364 369
365 if patience <= iter : 370 if patience <= iter :
366 print('iter (%i) is superior than patience(%i). break', iter, patience) 371 print('iter (%i) is superior than patience(%i). break', (iter, patience))
367 break 372 break
368 373
369 374
370 375
371 end_time = time.clock() 376 end_time = time.clock()
449 y = T.lvector() # the labels are presented as 1D vector of 454 y = T.lvector() # the labels are presented as 1D vector of
450 # [long int] labels 455 # [long int] labels
451 456
452 # construct the denoising autoencoder class 457 # construct the denoising autoencoder class
453 n_ins = 28*28 458 n_ins = 28*28
454 encoder = dA(n_ins, n_code_layer, input = x.reshape((batch_size,n_ins))) 459 encoder = dA(n_ins, n_code_layer, complexity, input = x.reshape((batch_size,n_ins)))
455 460
456 # Train autoencoder 461 # Train autoencoder
457 462
458 # compute gradients of the layer parameters 463 # compute gradients of the layer parameters
459 gW = T.grad(encoder.cost, encoder.W) 464 gW = T.grad(encoder.cost, encoder.W)