annotate code_tutoriel/DBN.py @ 451:227ebc0be7ae

Add a graph for the NIST training set and normalize the values.
author Arnaud Bergeron <abergeron@gmail.com>
date Mon, 10 May 2010 13:44:11 -0400
parents 4bc5eeec6394
children
rev   line source
165
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
1 """
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
2 """
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
3 import os
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
4
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
5 import numpy, time, cPickle, gzip
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
6
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
7 import theano
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
8 import theano.tensor as T
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
9 from theano.tensor.shared_randomstreams import RandomStreams
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
10
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
11 from logistic_sgd import LogisticRegression, load_data
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
12 from mlp import HiddenLayer
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
13 from rbm import RBM
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
14
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
15
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
16
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
17 class DBN(object):
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
18 """
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
19 """
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
20
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
21 def __init__(self, numpy_rng, theano_rng = None, n_ins = 784,
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
22 hidden_layers_sizes = [500,500], n_outs = 10):
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
23 """This class is made to support a variable number of layers.
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
24
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
25 :type numpy_rng: numpy.random.RandomState
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
26 :param numpy_rng: numpy random number generator used to draw initial
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
27 weights
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
28
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
29 :type theano_rng: theano.tensor.shared_randomstreams.RandomStreams
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
30 :param theano_rng: Theano random generator; if None is given one is
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
31 generated based on a seed drawn from `rng`
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
32
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
33 :type n_ins: int
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
34 :param n_ins: dimension of the input to the DBN
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
35
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
36 :type n_layers_sizes: list of ints
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
37 :param n_layers_sizes: intermidiate layers size, must contain
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
38 at least one value
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
39
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
40 :type n_outs: int
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
41 :param n_outs: dimension of the output of the network
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
42 """
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
43
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
44 self.sigmoid_layers = []
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
45 self.rbm_layers = []
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
46 self.params = []
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
47 self.n_layers = len(hidden_layers_sizes)
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
48
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
49 assert self.n_layers > 0
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
50
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
51 if not theano_rng:
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
52 theano_rng = RandomStreams(numpy_rng.randint(2**30))
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
53
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
54 # allocate symbolic variables for the data
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
55 self.x = T.matrix('x') # the data is presented as rasterized images
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
56 self.y = T.ivector('y') # the labels are presented as 1D vector of
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
57 # [int] labels
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
58
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
59 # The DBN is an MLP, for which all weights of intermidiate layers are shared with a
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
60 # different RBM. We will first construct the DBN as a deep multilayer perceptron, and
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
61 # when constructing each sigmoidal layer we also construct an RBM that shares weights
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
62 # with that layer. During pretraining we will train these RBMs (which will lead
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
63 # to chainging the weights of the MLP as well) During finetuning we will finish
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
64 # training the DBN by doing stochastic gradient descent on the MLP.
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
65
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
66 for i in xrange( self.n_layers ):
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
67 # construct the sigmoidal layer
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
68
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
69 # the size of the input is either the number of hidden units of the layer below or
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
70 # the input size if we are on the first layer
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
71 if i == 0 :
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
72 input_size = n_ins
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
73 else:
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
74 input_size = hidden_layers_sizes[i-1]
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
75
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
76 # the input to this layer is either the activation of the hidden layer below or the
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
77 # input of the DBN if you are on the first layer
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
78 if i == 0 :
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
79 layer_input = self.x
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
80 else:
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
81 layer_input = self.sigmoid_layers[-1].output
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
82
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
83 sigmoid_layer = HiddenLayer(rng = numpy_rng,
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
84 input = layer_input,
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
85 n_in = input_size,
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
86 n_out = hidden_layers_sizes[i],
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
87 activation = T.nnet.sigmoid)
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
88
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
89 # add the layer to our list of layers
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
90 self.sigmoid_layers.append(sigmoid_layer)
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
91
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
92 # its arguably a philosophical question... but we are going to only declare that
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
93 # the parameters of the sigmoid_layers are parameters of the DBN. The visible
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
94 # biases in the RBM are parameters of those RBMs, but not of the DBN.
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
95 self.params.extend(sigmoid_layer.params)
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
96
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
97 # Construct an RBM that shared weights with this layer
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
98 rbm_layer = RBM(numpy_rng = numpy_rng, theano_rng = theano_rng,
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
99 input = layer_input,
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
100 n_visible = input_size,
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
101 n_hidden = hidden_layers_sizes[i],
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
102 W = sigmoid_layer.W,
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
103 hbias = sigmoid_layer.b)
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
104 self.rbm_layers.append(rbm_layer)
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
105
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
106
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
107 # We now need to add a logistic layer on top of the MLP
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
108 self.logLayer = LogisticRegression(\
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
109 input = self.sigmoid_layers[-1].output,\
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
110 n_in = hidden_layers_sizes[-1], n_out = n_outs)
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
111 self.params.extend(self.logLayer.params)
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
112
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
113 # construct a function that implements one step of fine-tuning compute the cost for
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
114 # second phase of training, defined as the negative log likelihood
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
115 self.finetune_cost = self.logLayer.negative_log_likelihood(self.y)
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
116
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
117 # compute the gradients with respect to the model parameters
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
118 # symbolic variable that points to the number of errors made on the
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
119 # minibatch given by self.x and self.y
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
120 self.errors = self.logLayer.errors(self.y)
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
121
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
122 def pretraining_functions(self, train_set_x, batch_size):
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
123 ''' Generates a list of functions, for performing one step of gradient descent at a
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
124 given layer. The function will require as input the minibatch index, and to train an
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
125 RBM you just need to iterate, calling the corresponding function on all minibatch
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
126 indexes.
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
127
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
128 :type train_set_x: theano.tensor.TensorType
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
129 :param train_set_x: Shared var. that contains all datapoints used for training the RBM
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
130 :type batch_size: int
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
131 :param batch_size: size of a [mini]batch
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
132 '''
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
133
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
134 # index to a [mini]batch
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
135 index = T.lscalar('index') # index to a minibatch
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
136 learning_rate = T.scalar('lr') # learning rate to use
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
137
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
138 # number of batches
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
139 n_batches = train_set_x.value.shape[0] / batch_size
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
140 # begining of a batch, given `index`
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
141 batch_begin = index * batch_size
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
142 # ending of a batch given `index`
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
143 batch_end = batch_begin+batch_size
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
144
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
145 pretrain_fns = []
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
146 for rbm in self.rbm_layers:
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
147
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
148 # get the cost and the updates list
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
149 # TODO: change cost function to reconstruction error
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
150 cost,updates = rbm.cd(learning_rate, persistent=None)
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
151
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
152 # compile the theano function
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
153 fn = theano.function(inputs = [index,
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
154 theano.Param(learning_rate, default = 0.1)],
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
155 outputs = cost,
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
156 updates = updates,
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
157 givens = {self.x :train_set_x[batch_begin:batch_end]})
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
158 # append `fn` to the list of functions
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
159 pretrain_fns.append(fn)
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
160
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
161 return pretrain_fns
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
162
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
163
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
164 def build_finetune_functions(self, datasets, batch_size, learning_rate):
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
165 '''Generates a function `train` that implements one step of finetuning, a function
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
166 `validate` that computes the error on a batch from the validation set, and a function
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
167 `test` that computes the error on a batch from the testing set
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
168
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
169 :type datasets: list of pairs of theano.tensor.TensorType
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
170 :param datasets: It is a list that contain all the datasets; the has to contain three
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
171 pairs, `train`, `valid`, `test` in this order, where each pair is formed of two Theano
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
172 variables, one for the datapoints, the other for the labels
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
173 :type batch_size: int
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
174 :param batch_size: size of a minibatch
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
175 :type learning_rate: float
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
176 :param learning_rate: learning rate used during finetune stage
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
177 '''
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
178
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
179 (train_set_x, train_set_y) = datasets[0]
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
180 (valid_set_x, valid_set_y) = datasets[1]
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
181 (test_set_x , test_set_y ) = datasets[2]
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
182
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
183 # compute number of minibatches for training, validation and testing
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
184 n_valid_batches = valid_set_x.value.shape[0] / batch_size
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
185 n_test_batches = test_set_x.value.shape[0] / batch_size
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
186
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
187 index = T.lscalar('index') # index to a [mini]batch
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
188
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
189 # compute the gradients with respect to the model parameters
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
190 gparams = T.grad(self.finetune_cost, self.params)
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
191
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
192 # compute list of fine-tuning updates
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
193 updates = {}
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
194 for param, gparam in zip(self.params, gparams):
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
195 updates[param] = param - gparam*learning_rate
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
196
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
197 train_fn = theano.function(inputs = [index],
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
198 outputs = self.finetune_cost,
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
199 updates = updates,
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
200 givens = {
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
201 self.x : train_set_x[index*batch_size:(index+1)*batch_size],
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
202 self.y : train_set_y[index*batch_size:(index+1)*batch_size]})
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
203
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
204 test_score_i = theano.function([index], self.errors,
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
205 givens = {
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
206 self.x: test_set_x[index*batch_size:(index+1)*batch_size],
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
207 self.y: test_set_y[index*batch_size:(index+1)*batch_size]})
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
208
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
209 valid_score_i = theano.function([index], self.errors,
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
210 givens = {
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
211 self.x: valid_set_x[index*batch_size:(index+1)*batch_size],
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
212 self.y: valid_set_y[index*batch_size:(index+1)*batch_size]})
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
213
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
214 # Create a function that scans the entire validation set
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
215 def valid_score():
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
216 return [valid_score_i(i) for i in xrange(n_valid_batches)]
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
217
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
218 # Create a function that scans the entire test set
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
219 def test_score():
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
220 return [test_score_i(i) for i in xrange(n_test_batches)]
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
221
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
222 return train_fn, valid_score, test_score
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
223
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
224
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
225
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
226
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
227
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
228
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
229 def test_DBN( finetune_lr = 0.1, pretraining_epochs = 10, \
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
230 pretrain_lr = 0.1, training_epochs = 1000, \
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
231 dataset='mnist.pkl.gz'):
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
232 """
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
233 Demonstrates how to train and test a Deep Belief Network.
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
234
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
235 This is demonstrated on MNIST.
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
236
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
237 :type learning_rate: float
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
238 :param learning_rate: learning rate used in the finetune stage
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
239 :type pretraining_epochs: int
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
240 :param pretraining_epochs: number of epoch to do pretraining
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
241 :type pretrain_lr: float
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
242 :param pretrain_lr: learning rate to be used during pre-training
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
243 :type n_iter: int
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
244 :param n_iter: maximal number of iterations ot run the optimizer
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
245 :type dataset: string
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
246 :param dataset: path the the pickled dataset
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
247 """
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
248
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
249 print 'finetune_lr = ', finetune_lr
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
250 print 'pretrain_lr = ', pretrain_lr
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
251
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
252 datasets = load_data(dataset)
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
253
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
254 train_set_x, train_set_y = datasets[0]
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
255 valid_set_x, valid_set_y = datasets[1]
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
256 test_set_x , test_set_y = datasets[2]
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
257
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
258
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
259 batch_size = 20 # size of the minibatch
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
260
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
261 # compute number of minibatches for training, validation and testing
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
262 n_train_batches = train_set_x.value.shape[0] / batch_size
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
263
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
264 # numpy random generator
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
265 numpy_rng = numpy.random.RandomState(123)
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
266 print '... building the model'
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
267 # construct the Deep Belief Network
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
268 dbn = DBN(numpy_rng = numpy_rng, n_ins = 28*28,
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
269 hidden_layers_sizes = [1000,1000,1000],
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
270 n_outs = 10)
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
271
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
272
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
273 #########################
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
274 # PRETRAINING THE MODEL #
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
275 #########################
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
276 print '... getting the pretraining functions'
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
277 pretraining_fns = dbn.pretraining_functions(
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
278 train_set_x = train_set_x,
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
279 batch_size = batch_size )
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
280
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
281 print '... pre-training the model'
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
282 start_time = time.clock()
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
283 ## Pre-train layer-wise
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
284 for i in xrange(dbn.n_layers):
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
285 # go through pretraining epochs
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
286 for epoch in xrange(pretraining_epochs):
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
287 # go through the training set
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
288 c = []
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
289 for batch_index in xrange(n_train_batches):
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
290 c.append(pretraining_fns[i](index = batch_index,
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
291 lr = pretrain_lr ) )
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
292 print 'Pre-training layer %i, epoch %d, cost '%(i,epoch),numpy.mean(c)
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
293
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
294 end_time = time.clock()
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
295
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
296 print ('Pretraining took %f minutes' %((end_time-start_time)/60.))
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
297
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
298 ########################
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
299 # FINETUNING THE MODEL #
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
300 ########################
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
301
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
302 # get the training, validation and testing function for the model
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
303 print '... getting the finetuning functions'
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
304 train_fn, validate_model, test_model = dbn.build_finetune_functions (
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
305 datasets = datasets, batch_size = batch_size,
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
306 learning_rate = finetune_lr)
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
307
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
308 print '... finetunning the model'
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
309 # early-stopping parameters
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
310 patience = 10000 # look as this many examples regardless
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
311 patience_increase = 2. # wait this much longer when a new best is
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
312 # found
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
313 improvement_threshold = 0.995 # a relative improvement of this much is
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
314 # considered significant
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
315 validation_frequency = min(n_train_batches, patience/2)
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
316 # go through this many
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
317 # minibatche before checking the network
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
318 # on the validation set; in this case we
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
319 # check every epoch
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
320
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
321
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
322 best_params = None
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
323 best_validation_loss = float('inf')
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
324 test_score = 0.
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
325 start_time = time.clock()
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
326
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
327 done_looping = False
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
328 epoch = 0
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
329
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
330 while (epoch < training_epochs) and (not done_looping):
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
331 epoch = epoch + 1
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
332 for minibatch_index in xrange(n_train_batches):
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
333
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
334 minibatch_avg_cost = train_fn(minibatch_index)
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
335 iter = epoch * n_train_batches + minibatch_index
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
336
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
337 if (iter+1) % validation_frequency == 0:
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
338
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
339 validation_losses = validate_model()
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
340 this_validation_loss = numpy.mean(validation_losses)
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
341 print('epoch %i, minibatch %i/%i, validation error %f %%' % \
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
342 (epoch, minibatch_index+1, n_train_batches, \
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
343 this_validation_loss*100.))
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
344
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
345
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
346 # if we got the best validation score until now
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
347 if this_validation_loss < best_validation_loss:
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
348
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
349 #improve patience if loss improvement is good enough
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
350 if this_validation_loss < best_validation_loss * \
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
351 improvement_threshold :
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
352 patience = max(patience, iter * patience_increase)
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
353
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
354 # save best validation score and iteration number
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
355 best_validation_loss = this_validation_loss
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
356 best_iter = iter
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
357
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
358 # test it on the test set
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
359 test_losses = test_model()
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
360 test_score = numpy.mean(test_losses)
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
361 print((' epoch %i, minibatch %i/%i, test error of best '
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
362 'model %f %%') %
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
363 (epoch, minibatch_index+1, n_train_batches,
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
364 test_score*100.))
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
365
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
366
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
367 if patience <= iter :
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
368 done_looping = True
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
369 break
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
370
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
371 end_time = time.clock()
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
372 print(('Optimization complete with best validation score of %f %%,'
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
373 'with test performance %f %%') %
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
374 (best_validation_loss * 100., test_score*100.))
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
375 print ('The code ran for %f minutes' % ((end_time-start_time)/60.))
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
376
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
377
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
378
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
379
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
380
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
381 if __name__ == '__main__':
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
382 pretrain_lr = numpy.float(os.sys.argv[1])
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
383 finetune_lr = numpy.float(os.sys.argv[2])
4bc5eeec6394 Updating the tutorial code to the latest revisions.
Dumitru Erhan <dumitru.erhan@gmail.com>
parents:
diff changeset
384 test_DBN(pretrain_lr=pretrain_lr, finetune_lr=finetune_lr)