Mercurial > pylearn

annotate mlp.py @ 141:f5f235bebee4

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
...
author Frederic Bastien <bastienf@iro.umontreal.ca>
date Mon, 12 May 2008 14:13:39 -0400
parents 2ca8dccba270
children 4efe6d36c061
rev   line source
111
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
1
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
2 from learner import *
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
3 from theano import tensor as t
118
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
4 from nnet_ops import *
111
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
5
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
6 # this is one of the simplest example of learner, and illustrates
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
7 # the use of theano
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
8
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
9
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
10 class OneHiddenLayerNNetClassifier(MinibatchUpdatesTLearner):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
11 """
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
12 Implement a straightforward classicial feedforward
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
13 one-hidden-layer neural net, with L2 regularization.
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
14
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
15 The predictor parameters are obtained by minibatch/online gradient descent.
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
16 Training can proceed sequentially (with multiple calls to update with
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
17 different disjoint subsets of the training sets).
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
18
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
19 Hyper-parameters:
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
20 - L2_regularizer
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
21 - learning_rate
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
22 - n_hidden
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
23
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
24 For each (input_t,output_t) pair in a minibatch,::
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
25
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
26 output_activations_t = b2+W2*tanh(b1+W1*input_t)
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
27 output_t = softmax(output_activations_t)
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
28 output_class_t = argmax(output_activations_t)
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
29 class_error_t = 1_{output_class_t != target_t}
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
30 nll_t = -log(output_t[target_t])
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
31
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
32 and the training criterion is::
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
33
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
34 loss = L2_regularizer*(||W1||^2 + ||W2||^2) + sum_t nll_t
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
35
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
36 The parameters are [b1,W1,b2,W2] and are obtained by minimizing the loss by
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
37 stochastic minibatch gradient descent::
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
38
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
39 parameters[i] -= learning_rate * dloss/dparameters[i]
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
40
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
41 The fields and attributes expected and produced by use and update are the following:
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
42
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
43 - Input and output fields (example-wise quantities):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
44
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
45 - 'input' (always expected by use and update)
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
46 - 'target' (optionally expected by use and always by update)
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
47 - 'output' (optionally produced by use)
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
48 - 'output_class' (optionally produced by use)
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
49 - 'class_error' (optionally produced by use)
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
50 - 'nll' (optionally produced by use)
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
51
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
52 - optional attributes (optionally expected as input_dataset attributes)
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
53 (warning, this may be dangerous, the 'use' method will use those provided in the
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
54 input_dataset rather than those learned during 'update'; currently no support
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
55 for providing these to update):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
56
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
57 - 'L2_regularizer'
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
58 - 'b1'
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
59 - 'W1'
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
60 - 'b2'
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
61 - 'W2'
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
62 - 'parameters' = [b1, W1, b2, W2]
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
63 - 'regularization_term'
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
64
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
65 """
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
66
121
2ca8dccba270 debugging mlp.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 118
diff changeset
67 def __init__(self,n_hidden,n_classes,learning_rate,init_range=1.):
2ca8dccba270 debugging mlp.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 118
diff changeset
68 self._n_outputs = n_classes
2ca8dccba270 debugging mlp.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 118
diff changeset
69 self._n_hidden = n_hidden
2ca8dccba270 debugging mlp.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 118
diff changeset
70 self._init_range = init_range
2ca8dccba270 debugging mlp.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 118
diff changeset
71 self.learning_rate = learning_rate # this is the float
2ca8dccba270 debugging mlp.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 118
diff changeset
72 self._learning_rate = t.scalar('learning_rate') # this is the symbol
2ca8dccba270 debugging mlp.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 118
diff changeset
73 self._input = t.matrix('input') # n_examples x n_inputs
2ca8dccba270 debugging mlp.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 118
diff changeset
74 self._target = t.matrix('target','int32') # n_examples x n_outputs
2ca8dccba270 debugging mlp.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 118
diff changeset
75 self._L2_regularizer = t.scalar('L2_regularizer')
2ca8dccba270 debugging mlp.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 118
diff changeset
76 self._W1 = t.matrix('W1')
2ca8dccba270 debugging mlp.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 118
diff changeset
77 self._W2 = t.matrix('W2')
2ca8dccba270 debugging mlp.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 118
diff changeset
78 self._b1 = t.row('b1')
2ca8dccba270 debugging mlp.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 118
diff changeset
79 self._b2 = t.row('b2')
2ca8dccba270 debugging mlp.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 118
diff changeset
80 self._regularization_term = self._L2_regularizer * (t.dot(self._W1,self._W1) + t.dot(self._W2,self._W2))
2ca8dccba270 debugging mlp.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 118
diff changeset
81 self._output_activations =self._b2+t.dot(t.tanh(self._b1+t.dot(self._input,self._W1.T)),self._W2.T)
2ca8dccba270 debugging mlp.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 118
diff changeset
82 self._nll,self._output = crossentropy_softmax_1hot(self._output_activations,self._target)
2ca8dccba270 debugging mlp.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 118
diff changeset
83 self._output_class = t.argmax(self._output,1)
2ca8dccba270 debugging mlp.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 118
diff changeset
84 self._class_error = self._output_class != self._target
2ca8dccba270 debugging mlp.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 118
diff changeset
85 self._minibatch_criterion = self._nll + self._regularization_term / t.shape(self._input)[0]
2ca8dccba270 debugging mlp.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 118
diff changeset
86 MinibatchUpdatesTLearner.__init__(self)
2ca8dccba270 debugging mlp.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 118
diff changeset
87
111
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
88 def attributeNames(self):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
89 return ["parameters","b1","W2","b2","W2", "L2_regularizer","regularization_term"]
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
90
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
91 def parameterAttributes(self):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
92 return ["b1","W1", "b2", "W2"]
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
93
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
94 def useInputAttributes(self):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
95 return self.parameterAttributes()
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
96
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
97 def useOutputAttributes(self):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
98 return []
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
99
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
100 def updateInputAttributes(self):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
101 return self.parameterAttributes() + ["L2_regularizer"]
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
102
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
103 def updateMinibatchInputFields(self):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
104 return ["input","target"]
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
105
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
106 def updateEndOutputAttributes(self):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
107 return ["regularization_term"]
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
108
118
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
109 def lossAttribute(self):
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
110 return "minibatch_criterion"
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
111
111
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
112 def defaultOutputFields(self, input_fields):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
113 output_fields = ["output", "output_class",]
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
114 if "target" in input_fields:
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
115 output_fields += ["class_error", "nll"]
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
116 return output_fields
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
117
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
118 def allocate(self,minibatch):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
119 minibatch_n_inputs = minibatch["input"].shape[1]
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
120 if not self._n_inputs:
118
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
121 self._n_inputs = minibatch_n_inputs
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
122 self.b1 = numpy.zeros(self._n_hidden)
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
123 self.b2 = numpy.zeros(self._n_outputs)
111
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
124 self.forget()
118
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
125 elif self._n_inputs!=minibatch_n_inputs:
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
126 # if the input changes dimension on the fly, we resize and forget everything
111
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
127 self.forget()
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
128
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
129 def forget(self):
118
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
130 if self._n_inputs:
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
131 r = self._init_range/math.sqrt(self._n_inputs)
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
132 self.W1 = numpy.random.uniform(low=-r,high=r,
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
133 size=(self._n_hidden,self._n_inputs))
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
134 r = self._init_range/math.sqrt(self._n_hidden)
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
135 self.W2 = numpy.random.uniform(low=-r,high=r,
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
136 size=(self._n_outputs,self._n_hidden))
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
137 self.b1[:]=0
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
138 self.b2[:]=0
111
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
139
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
140
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
141 class MLP(MinibatchUpdatesTLearner):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
142 """
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
143 Implement a feedforward multi-layer perceptron, with or without L1 and/or L2 regularization.
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
144
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
145 The predictor parameters are obtained by minibatch/online gradient descent.
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
146 Training can proceed sequentially (with multiple calls to update with
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
147 different disjoint subsets of the training sets).
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
148
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
149 Hyper-parameters:
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
150 - L1_regularizer
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
151 - L2_regularizer
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
152 - neuron_sparsity_regularizer
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
153 - initial_learning_rate
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
154 - learning_rate_decrease_rate
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
155 - n_hidden_per_layer (a list of integers)
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
156 - activation_function ("sigmoid","tanh", or "ratio")
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
157
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
158 The output/task type (classification, regression, etc.) is obtained by specializing MLP.
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
159
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
160 For each (input[t],output[t]) pair in a minibatch,::
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
161
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
162 activation[0] = input_t
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
163 for k=1 to n_hidden_layers:
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
164 activation[k]=activation_function(b[k]+ W[k]*activation[k-1])
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
165 output_t = output_activation_function(b[n_hidden_layers+1]+W[n_hidden_layers+1]*activation[n_hidden_layers])
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
166
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
167 and the b and W are obtained by minimizing the following by stochastic minibatch gradient descent::
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
168
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
169 L2_regularizer sum_{ijk} W_{kij}^2 + L1_regularizer sum_{kij} |W_{kij}|
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
170 + neuron_sparsity_regularizer sum_{ki} |b_{ki} + infinity|
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
171 - sum_t log P_{output_model}(target_t | output_t)
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
172
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
173 The fields and attributes expected and produced by use and update are the following:
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
174
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
175 - Input and output fields (example-wise quantities):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
176
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
177 - 'input' (always expected by use and update)
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
178 - 'target' (optionally expected by use and always by update)
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
179 - 'output' (optionally produced by use)
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
180 - error fields produced by sub-class of MLP
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
181
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
182 - optional attributes (optionally expected as input_dataset attributes)
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
183 (warning, this may be dangerous, the 'use' method will use those provided in the
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
184 input_dataset rather than those learned during 'update'; currently no support
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
185 for providing these to update):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
186
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
187 - 'L1_regularizer'
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
188 - 'L2_regularizer'
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
189 - 'b'
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
190 - 'W'
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
191 - 'parameters' = [b[1], W[1], b[2], W[2], ...]
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
192 - 'regularization_term'
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
193
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
194 """
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
195
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
196 def attributeNames(self):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
197 return ["parameters","b","W","L1_regularizer","L2_regularizer","neuron_sparsity_regularizer","regularization_term"]
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
198
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
199 def useInputAttributes(self):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
200 return ["b","W"]
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
201
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
202 def useOutputAttributes(self):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
203 return []
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
204
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
205 def updateInputAttributes(self):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
206 return ["b","W","L1_regularizer","L2_regularizer","neuron_sparsity_regularizer"]
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
207
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
208 def updateMinibatchInputFields(self):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
209 return ["input","target"]
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
210
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
211 def updateMinibatchInputAttributes(self):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
212 return ["b","W"]
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
213
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
214 def updateMinibatchOutputAttributes(self):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
215 return ["new_XtX","new_XtY"]
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
216
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
217 def updateEndInputAttributes(self):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
218 return ["theta","XtX","XtY"]
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
219
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
220 def updateEndOutputAttributes(self):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
221 return ["new_theta","b","W","regularization_term"] # CHECK: WILL b AND W CONTAIN OLD OR NEW THETA? @todo i.e. order of computation = ?
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
222
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
223 def parameterAttributes(self):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
224 return ["b","W"]
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
225
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
226 def defaultOutputFields(self, input_fields):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
227 output_fields = ["output"]
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
228 if "target" in input_fields:
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
229 output_fields.append("squared_error")
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
230 return output_fields
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
231
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
232 def __init__(self):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
233 self._input = t.matrix('input') # n_examples x n_inputs
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
234 self._target = t.matrix('target') # n_examples x n_outputs
121
2ca8dccba270 debugging mlp.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 118
diff changeset
235 self._L2_regularizer = t.scalar('L2_regularizer')
111
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
236 self._theta = t.matrix('theta')
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
237 self._W = self._theta[:,1:]
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
238 self._b = self._theta[:,0]
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
239 self._XtX = t.matrix('XtX')
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
240 self._XtY = t.matrix('XtY')
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
241 self._extended_input = t.prepend_one_to_each_row(self._input)
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
242 self._output = t.dot(self._input,self._W.T) + self._b # (n_examples , n_outputs) matrix
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
243 self._squared_error = t.sum_within_rows(t.sqr(self._output-self._target)) # (n_examples ) vector
118
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
244 self._regularizer = self._L2_regularizer * t.dot(self._W,self._W)
111
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
245 self._new_XtX = add_inplace(self._XtX,t.dot(self._extended_input.T,self._extended_input))
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
246 self._new_XtY = add_inplace(self._XtY,t.dot(self._extended_input.T,self._target))
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
247 self._new_theta = t.solve_inplace(self._theta,self._XtX,self._XtY)
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
248
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
249 OneShotTLearner.__init__(self)
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
250
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
251 def allocate(self,minibatch):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
252 minibatch_n_inputs = minibatch["input"].shape[1]
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
253 minibatch_n_outputs = minibatch["target"].shape[1]
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
254 if not self._n_inputs:
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
255 self._n_inputs = minibatch_n_inputs
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
256 self._n_outputs = minibatch_n_outputs
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
257 self.XtX = numpy.zeros((1+self._n_inputs,1+self._n_inputs))
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
258 self.XtY = numpy.zeros((1+self._n_inputs,self._n_outputs))
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
259 self.theta = numpy.zeros((self._n_outputs,1+self._n_inputs))
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
260 self.forget()
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
261 elif self._n_inputs!=minibatch_n_inputs or self._n_outputs!=minibatch_n_outputs:
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
262 # if the input or target changes dimension on the fly, we resize and forget everything
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
263 self.forget()
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
264
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
265 def forget(self):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
266 if self._n_inputs and self._n_outputs:
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
267 self.XtX.resize((1+self.n_inputs,1+self.n_inputs))
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
268 self.XtY.resize((1+self.n_inputs,self.n_outputs))
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
269 self.XtX.data[:,:]=0
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
270 self.XtY.data[:,:]=0
118
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
271 numpy.diag(self.XtX.data)[1:]=self.L2_regularizer
111
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
diff changeset
272