Mercurial > pylearn
annotate learner.py @ 147:625d2b21ee48
Automated merge with ssh://p-omega1@lgcm.iro.umontreal.ca/tlearn
author | Yoshua Bengio <bengioy@iro.umontreal.ca> |
---|---|
date | Mon, 12 May 2008 15:50:34 -0400 |
parents | ceae4de18981 |
children | 4803cb76e26b |
rev | line source |
---|---|
1
2cd82666b9a7
Added statscollector and started writing dataset and learner.
bengioy@esprit.iro.umontreal.ca
parents:
0
diff
changeset
|
1 |
134
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
2 from dataset import AttributesHolder,AbstractFunction,ApplyFunctionDataSet,DataSet,CachedDataSet |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
3 import theano |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
4 from theano import compile |
129
4c2280edcaf5
Fixed typos in learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
128
diff
changeset
|
5 from theano import tensor as t |
1
2cd82666b9a7
Added statscollector and started writing dataset and learner.
bengioy@esprit.iro.umontreal.ca
parents:
0
diff
changeset
|
6 |
110
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
7 class Learner(AttributesHolder): |
132
f6505ec32dc3
Updated documentation slightly
Joseph Turian <turian@gmail.com>
parents:
131
diff
changeset
|
8 """ |
f6505ec32dc3
Updated documentation slightly
Joseph Turian <turian@gmail.com>
parents:
131
diff
changeset
|
9 Base class for learning algorithms, provides an interface |
1
2cd82666b9a7
Added statscollector and started writing dataset and learner.
bengioy@esprit.iro.umontreal.ca
parents:
0
diff
changeset
|
10 that allows various algorithms to be applicable to generic learning |
2cd82666b9a7
Added statscollector and started writing dataset and learner.
bengioy@esprit.iro.umontreal.ca
parents:
0
diff
changeset
|
11 algorithms. |
2cd82666b9a7
Added statscollector and started writing dataset and learner.
bengioy@esprit.iro.umontreal.ca
parents:
0
diff
changeset
|
12 |
132
f6505ec32dc3
Updated documentation slightly
Joseph Turian <turian@gmail.com>
parents:
131
diff
changeset
|
13 A L{Learner} can be seen as a learning algorithm, a function that when |
135
0d8e721cc63c
Fixed bugs in dataset to make test_mlp.py work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
134
diff
changeset
|
14 applied to training data returns a learned function (which is an object that |
0d8e721cc63c
Fixed bugs in dataset to make test_mlp.py work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
134
diff
changeset
|
15 can be applied to other data and return some output data). |
0d8e721cc63c
Fixed bugs in dataset to make test_mlp.py work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
134
diff
changeset
|
16 |
1
2cd82666b9a7
Added statscollector and started writing dataset and learner.
bengioy@esprit.iro.umontreal.ca
parents:
0
diff
changeset
|
17 """ |
2cd82666b9a7
Added statscollector and started writing dataset and learner.
bengioy@esprit.iro.umontreal.ca
parents:
0
diff
changeset
|
18 |
10
80bf5492e571
Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents:
1
diff
changeset
|
19 def __init__(self): |
80bf5492e571
Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents:
1
diff
changeset
|
20 pass |
80bf5492e571
Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents:
1
diff
changeset
|
21 |
80bf5492e571
Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents:
1
diff
changeset
|
22 def forget(self): |
80bf5492e571
Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents:
1
diff
changeset
|
23 """ |
80bf5492e571
Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents:
1
diff
changeset
|
24 Reset the state of the learner to a blank slate, before seeing |
80bf5492e571
Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents:
1
diff
changeset
|
25 training data. The operation may be non-deterministic if the |
80bf5492e571
Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents:
1
diff
changeset
|
26 learner has a random number generator that is set to use a |
80bf5492e571
Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents:
1
diff
changeset
|
27 different seed each time it forget() is called. |
80bf5492e571
Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents:
1
diff
changeset
|
28 """ |
80bf5492e571
Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents:
1
diff
changeset
|
29 raise NotImplementedError |
80bf5492e571
Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents:
1
diff
changeset
|
30 |
14 | 31 def update(self,training_set,train_stats_collector=None): |
10
80bf5492e571
Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents:
1
diff
changeset
|
32 """ |
80bf5492e571
Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents:
1
diff
changeset
|
33 Continue training a learner, with the evidence provided by the given training set. |
80bf5492e571
Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents:
1
diff
changeset
|
34 Hence update can be called multiple times. This is particularly useful in the |
80bf5492e571
Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents:
1
diff
changeset
|
35 on-line setting or the sequential (Bayesian or not) settings. |
80bf5492e571
Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents:
1
diff
changeset
|
36 The result is a function that can be applied on data, with the same |
80bf5492e571
Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents:
1
diff
changeset
|
37 semantics of the Learner.use method. |
75
90e4c0784d6e
Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
20
diff
changeset
|
38 |
132
f6505ec32dc3
Updated documentation slightly
Joseph Turian <turian@gmail.com>
parents:
131
diff
changeset
|
39 The user may optionally provide a training L{StatsCollector} that is used to record |
75
90e4c0784d6e
Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
20
diff
changeset
|
40 some statistics of the outputs computed during training. It is update(d) during |
90e4c0784d6e
Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
20
diff
changeset
|
41 training. |
10
80bf5492e571
Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents:
1
diff
changeset
|
42 """ |
13
633453635d51
Starting to work on gradient_based_learner.py
bengioy@bengiomac.local
parents:
10
diff
changeset
|
43 return self.use # default behavior is 'non-adaptive', i.e. update does not do anything |
10
80bf5492e571
Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents:
1
diff
changeset
|
44 |
80bf5492e571
Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents:
1
diff
changeset
|
45 |
20
266c68cb6136
Minor editions, plus adding untested ApplyFunctionDataset for GradientLearner in the works.
bengioy@bengiomac.local
parents:
14
diff
changeset
|
46 def __call__(self,training_set,train_stats_collector=None): |
10
80bf5492e571
Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents:
1
diff
changeset
|
47 """ |
80bf5492e571
Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents:
1
diff
changeset
|
48 Train a learner from scratch using the provided training set, |
80bf5492e571
Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents:
1
diff
changeset
|
49 and return the learned function. |
80bf5492e571
Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents:
1
diff
changeset
|
50 """ |
80bf5492e571
Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents:
1
diff
changeset
|
51 self.forget() |
133 | 52 return self.update(training_set,train_stats_collector) |
1
2cd82666b9a7
Added statscollector and started writing dataset and learner.
bengioy@esprit.iro.umontreal.ca
parents:
0
diff
changeset
|
53 |
128 | 54 def use(self,input_dataset,output_fieldnames=None, |
135
0d8e721cc63c
Fixed bugs in dataset to make test_mlp.py work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
134
diff
changeset
|
55 test_stats_collector=None,copy_inputs=False, |
128 | 56 put_stats_in_output_dataset=True, |
57 output_attributes=[]): | |
58 """ | |
132
f6505ec32dc3
Updated documentation slightly
Joseph Turian <turian@gmail.com>
parents:
131
diff
changeset
|
59 Once a L{Learner} has been trained by one or more call to 'update', it can |
f6505ec32dc3
Updated documentation slightly
Joseph Turian <turian@gmail.com>
parents:
131
diff
changeset
|
60 be used with one or more calls to 'use'. The argument is an input L{DataSet} (possibly |
f6505ec32dc3
Updated documentation slightly
Joseph Turian <turian@gmail.com>
parents:
131
diff
changeset
|
61 containing a single example) and the result is an output L{DataSet} of the same length. |
128 | 62 If output_fieldnames is specified, it may be use to indicate which fields should |
132
f6505ec32dc3
Updated documentation slightly
Joseph Turian <turian@gmail.com>
parents:
131
diff
changeset
|
63 be constructed in the output L{DataSet} (for example ['output','classification_error']). |
128 | 64 Otherwise, self.defaultOutputFields is called to choose the output fields. |
20
266c68cb6136
Minor editions, plus adding untested ApplyFunctionDataset for GradientLearner in the works.
bengioy@bengiomac.local
parents:
14
diff
changeset
|
65 Optionally, if copy_inputs, the input fields (of the input_dataset) can be made |
132
f6505ec32dc3
Updated documentation slightly
Joseph Turian <turian@gmail.com>
parents:
131
diff
changeset
|
66 visible in the output L{DataSet} returned by this method. |
128 | 67 Optionally, attributes of the learner can be copied in the output dataset, |
68 and statistics computed by the stats collector also put in the output dataset. | |
69 Note the distinction between fields (which are example-wise quantities, e.g. 'input') | |
70 and attributes (which are not, e.g. 'regularization_term'). | |
71 | |
72 We provide here a default implementation that does all this using | |
73 a sub-class defined method: minibatchwiseUseFunction. | |
74 | |
75 @todo check if some of the learner attributes are actually SPECIFIED | |
76 as attributes of the input_dataset, and if so use their values instead | |
77 of the ones in the learner. | |
78 | |
79 The learner tries to compute in the output dataset the output fields specified. | |
80 If None is specified then self.defaultOutputFields(input_dataset.fieldNames()) | |
81 is called to determine the output fields. | |
82 | |
83 Attributes of the learner can also optionally be copied into the output dataset. | |
84 If output_attributes is None then all of the attributes in self.AttributeNames() | |
85 are copied in the output dataset, but if it is [] (the default), then none are copied. | |
86 If a test_stats_collector is provided, then its attributes (test_stats_collector.AttributeNames()) | |
87 are also copied into the output dataset attributes. | |
10
80bf5492e571
Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents:
1
diff
changeset
|
88 """ |
135
0d8e721cc63c
Fixed bugs in dataset to make test_mlp.py work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
134
diff
changeset
|
89 input_fieldnames = input_dataset.fieldNames() |
0d8e721cc63c
Fixed bugs in dataset to make test_mlp.py work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
134
diff
changeset
|
90 if not output_fieldnames: |
0d8e721cc63c
Fixed bugs in dataset to make test_mlp.py work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
134
diff
changeset
|
91 output_fieldnames = self.defaultOutputFields(input_fieldnames) |
0d8e721cc63c
Fixed bugs in dataset to make test_mlp.py work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
134
diff
changeset
|
92 |
0d8e721cc63c
Fixed bugs in dataset to make test_mlp.py work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
134
diff
changeset
|
93 minibatchwise_use_function = self.minibatchwiseUseFunction(input_fieldnames, |
128 | 94 output_fieldnames, |
95 test_stats_collector) | |
96 virtual_output_dataset = ApplyFunctionDataSet(input_dataset, | |
97 minibatchwise_use_function, | |
135
0d8e721cc63c
Fixed bugs in dataset to make test_mlp.py work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
134
diff
changeset
|
98 output_fieldnames, |
128 | 99 True,DataSet.numpy_vstack, |
100 DataSet.numpy_hstack) | |
101 # actually force the computation | |
102 output_dataset = CachedDataSet(virtual_output_dataset,True) | |
103 if copy_inputs: | |
104 output_dataset = input_dataset | output_dataset | |
105 # copy the wanted attributes in the dataset | |
106 if output_attributes is None: | |
107 output_attributes = self.attributeNames() | |
108 if output_attributes: | |
109 assert set(attribute_names) <= set(self.attributeNames()) | |
110 output_dataset.setAttributes(output_attributes, | |
111 self.names2attributes(output_attributes,return_copy=True)) | |
112 if test_stats_collector: | |
113 test_stats_collector.update(output_dataset) | |
114 if put_stats_in_output_dataset: | |
115 output_dataset.setAttributes(test_stats_collector.attributeNames(), | |
116 test_stats_collector.attributes()) | |
117 return output_dataset | |
77
1e2bb5bad636
toying with different ways to implement learners
bengioy@bengiomac.local
parents:
75
diff
changeset
|
118 |
128 | 119 def minibatchwiseUseFunction(self, input_fields, output_fields, stats_collector): |
120 """ | |
121 Returns a function that can map the given input fields to the given output fields | |
122 and to the attributes that the stats collector needs for its computation. | |
123 That function is expected to operate on minibatches. | |
124 The function returned makes use of the self.useInputAttributes() and | |
125 sets the attributes specified by self.useOutputAttributes(). | |
126 """ | |
143
b7ca3545186b
added missing raise
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
134
diff
changeset
|
127 raise AbstractFunction() |
b7ca3545186b
added missing raise
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
134
diff
changeset
|
128 |
92
c4726e19b8ec
Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
78
diff
changeset
|
129 def attributeNames(self): |
78 | 130 """ |
131 A Learner may have attributes that it wishes to export to other objects. To automate | |
132 such export, sub-classes should define here the names (list of strings) of these attributes. | |
107
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
133 |
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
134 @todo By default, attributeNames looks for all dictionary entries whose name does not start with _. |
78 | 135 """ |
136 return [] | |
77
1e2bb5bad636
toying with different ways to implement learners
bengioy@bengiomac.local
parents:
75
diff
changeset
|
137 |
128 | 138 def attributes(self,return_copy=False): |
139 """ | |
140 Return a list with the values of the learner's attributes (or optionally, a deep copy). | |
141 """ | |
142 return self.names2attributes(self.attributeNames(),return_copy) | |
143 | |
134
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
144 def names2attributes(self,names): |
128 | 145 """ |
146 Private helper function that maps a list of attribute names to a list | |
147 of (optionally copies) values of attributes. | |
148 """ | |
134
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
149 res=[] |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
150 for name in names: |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
151 assert name in names |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
152 res.append(self.__getattribute__(name)) |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
153 return res |
110
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
154 |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
155 def useInputAttributes(self): |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
156 """ |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
157 A subset of self.attributeNames() which are the names of attributes needed by use() in order |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
158 to do its work. |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
159 """ |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
160 raise AbstractFunction() |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
161 |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
162 def useOutputAttributes(self): |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
163 """ |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
164 A subset of self.attributeNames() which are the names of attributes modified/created by use() in order |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
165 to do its work. |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
166 """ |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
167 raise AbstractFunction() |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
168 |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
169 |
78 | 170 class TLearner(Learner): |
171 """ | |
172 TLearner is a virtual class of Learners that attempts to factor out of the definition | |
173 of a learner the steps that are common to many implementations of learning algorithms, | |
107
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
174 so as to leave only 'the equations' to define in particular sub-classes, using Theano. |
78 | 175 |
176 In the default implementations of use and update, it is assumed that the 'use' and 'update' methods | |
177 visit examples in the input dataset sequentially. In the 'use' method only one pass through the dataset is done, | |
178 whereas the sub-learner may wish to iterate over the examples multiple times. Subclasses where this | |
179 basic model is not appropriate can simply redefine update or use. | |
180 | |
181 Sub-classes must provide the following functions and functionalities: | |
182 - attributeNames(): defines all the names of attributes which can be used as fields or | |
183 attributes in input/output datasets or in stats collectors. | |
184 All these attributes are expected to be theano.Result objects | |
185 (with a .data property and recognized by theano.Function for compilation). | |
186 The sub-class constructor defines the relations between | |
187 the Theano variables that may be used by 'use' and 'update' | |
188 or by a stats collector. | |
189 - defaultOutputFields(input_fields): return a list of default dataset output fields when | |
190 None are provided by the caller of use. | |
92
c4726e19b8ec
Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
78
diff
changeset
|
191 The following naming convention is assumed and important. |
c4726e19b8ec
Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
78
diff
changeset
|
192 Attributes whose names are listed in attributeNames() can be of any type, |
c4726e19b8ec
Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
78
diff
changeset
|
193 but those that can be referenced as input/output dataset fields or as |
c4726e19b8ec
Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
78
diff
changeset
|
194 output attributes in 'use' or as input attributes in the stats collector |
c4726e19b8ec
Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
78
diff
changeset
|
195 should be associated with a Theano Result variable. If the exported attribute |
c4726e19b8ec
Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
78
diff
changeset
|
196 name is <name>, the corresponding Result name (an internal attribute of |
c4726e19b8ec
Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
78
diff
changeset
|
197 the TLearner, created in the sub-class constructor) should be _<name>. |
c4726e19b8ec
Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
78
diff
changeset
|
198 Typically <name> will be numpy ndarray and _<name> will be the corresponding |
c4726e19b8ec
Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
78
diff
changeset
|
199 Theano Tensor (for symbolic manipulation). |
107
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
200 |
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
201 @todo pousser dans Learner toute la poutine qui peut l'etre sans etre |
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
202 dependant de Theano |
78 | 203 """ |
92
c4726e19b8ec
Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
78
diff
changeset
|
204 |
c4726e19b8ec
Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
78
diff
changeset
|
205 def __init__(self): |
c4726e19b8ec
Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
78
diff
changeset
|
206 Learner.__init__(self) |
134
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
207 self.use_functions_dictionary={} |
110
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
208 |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
209 def defaultOutputFields(self, input_fields): |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
210 """ |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
211 Return a default list of output field names (to put in the output dataset). |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
212 This will be used when None are provided (as output_fields) by the caller of the 'use' method. |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
213 This may involve looking at the input_fields (names) available in the |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
214 input_dataset. |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
215 """ |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
216 raise AbstractFunction() |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
217 |
128 | 218 def minibatchwiseUseFunction(self, input_fields, output_fields, stats_collector): |
110
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
219 """ |
128 | 220 Implement minibatchwiseUseFunction by exploiting Theano compilation |
221 and the expression graph defined by a sub-class constructor. | |
92
c4726e19b8ec
Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
78
diff
changeset
|
222 """ |
c4726e19b8ec
Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
78
diff
changeset
|
223 if stats_collector: |
110
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
224 stats_collector_inputs = stats_collector.input2UpdateAttributes() |
92
c4726e19b8ec
Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
78
diff
changeset
|
225 for attribute in stats_collector_inputs: |
c4726e19b8ec
Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
78
diff
changeset
|
226 if attribute not in input_fields: |
c4726e19b8ec
Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
78
diff
changeset
|
227 output_fields.append(attribute) |
134
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
228 key = (tuple(input_fields),tuple(output_fields)) |
92
c4726e19b8ec
Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
78
diff
changeset
|
229 if key not in self.use_functions_dictionary: |
110
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
230 use_input_attributes = self.useInputAttributes() |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
231 use_output_attributes = self.useOutputAttributes() |
118
d0a1bd0378c6
Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
111
diff
changeset
|
232 complete_f = compile.function(self.names2OpResults(input_fields+use_input_attributes), |
d0a1bd0378c6
Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
111
diff
changeset
|
233 self.names2OpResults(output_fields+use_output_attributes)) |
110
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
234 def f(*input_field_values): |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
235 input_attribute_values = self.names2attributes(use_input_attributes) |
134
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
236 results = complete_f(*(list(input_field_values) + input_attribute_values)) |
110
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
237 output_field_values = results[0:len(output_fields)] |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
238 output_attribute_values = results[len(output_fields):len(results)] |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
239 if use_output_attributes: |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
240 self.setAttributes(use_output_attributes,output_attribute_values) |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
241 return output_field_values |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
242 self.use_functions_dictionary[key]=f |
92
c4726e19b8ec
Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
78
diff
changeset
|
243 return self.use_functions_dictionary[key] |
c4726e19b8ec
Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
78
diff
changeset
|
244 |
110
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
245 def names2OpResults(self,names): |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
246 """ |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
247 Private helper function that maps a list of attribute names to a list |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
248 of corresponding Op Results (with the same name but with a '_' prefix). |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
249 """ |
133 | 250 return [self.__getattribute__('_'+name) for name in names] |
92
c4726e19b8ec
Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
78
diff
changeset
|
251 |
107
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
252 |
111
88257dfedf8c
Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
110
diff
changeset
|
253 class MinibatchUpdatesTLearner(TLearner): |
107
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
254 """ |
132
f6505ec32dc3
Updated documentation slightly
Joseph Turian <turian@gmail.com>
parents:
131
diff
changeset
|
255 This adds to L{TLearner} a |
110
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
256 - updateStart(), updateEnd(), updateMinibatch(minibatch), isLastEpoch(): |
107
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
257 functions executed at the beginning, the end, in the middle |
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
258 (for each minibatch) of the update method, and at the end |
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
259 of each epoch. This model only |
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
260 works for 'online' or one-shot learning that requires |
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
261 going only once through the training data. For more complicated |
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
262 models, more specialized subclasses of TLearner should be used |
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
263 or a learning-algorithm specific update method should be defined. |
111
88257dfedf8c
Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
110
diff
changeset
|
264 |
88257dfedf8c
Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
110
diff
changeset
|
265 - a 'parameters' attribute which is a list of parameters (whose names are |
88257dfedf8c
Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
110
diff
changeset
|
266 specified by the user's subclass with the parameterAttributes() method) |
88257dfedf8c
Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
110
diff
changeset
|
267 |
107
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
268 """ |
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
269 |
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
270 def __init__(self): |
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
271 TLearner.__init__(self) |
134
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
272 self.update_minibatch_function = compile.function(self.names2OpResults(self.updateMinibatchOutputAttributes()+ |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
273 self.updateMinibatchInputFields()), |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
274 self.names2OpResults(self.updateMinibatchOutputAttributes())) |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
275 self.update_end_function = compile.function(self.names2OpResults(self.updateEndInputAttributes()), |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
276 self.names2OpResults(self.updateEndOutputAttributes())) |
110
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
277 |
128 | 278 def allocate(self, minibatch): |
279 """ | |
132
f6505ec32dc3
Updated documentation slightly
Joseph Turian <turian@gmail.com>
parents:
131
diff
changeset
|
280 This function is called at the beginning of each L{updateMinibatch} |
128 | 281 and should be used to check that all required attributes have been |
282 allocated and initialized (usually this function calls forget() | |
283 when it has to do an initialization). | |
284 """ | |
285 raise AbstractFunction() | |
286 | |
110
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
287 def updateMinibatchInputFields(self): |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
288 raise AbstractFunction() |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
289 |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
290 def updateMinibatchInputAttributes(self): |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
291 raise AbstractFunction() |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
292 |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
293 def updateMinibatchOutputAttributes(self): |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
294 raise AbstractFunction() |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
295 |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
296 def updateEndInputAttributes(self): |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
297 raise AbstractFunction() |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
298 |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
299 def updateEndOutputAttributes(self): |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
300 raise AbstractFunction() |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
301 |
111
88257dfedf8c
Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
110
diff
changeset
|
302 def parameterAttributes(self): |
88257dfedf8c
Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
110
diff
changeset
|
303 raise AbstractFunction() |
88257dfedf8c
Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
110
diff
changeset
|
304 |
133 | 305 def updateStart(self,training_set): |
306 pass | |
110
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
307 |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
308 def updateEnd(self): |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
309 self.setAttributes(self.updateEndOutputAttributes(), |
134
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
310 self.update_end_function(*self.names2attributes(self.updateEndInputAttributes()))) |
111
88257dfedf8c
Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
110
diff
changeset
|
311 self.parameters = self.names2attributes(self.parameterAttributes()) |
107
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
312 |
110
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
313 def updateMinibatch(self,minibatch): |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
314 # make sure all required fields are allocated and initialized |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
315 self.allocate(minibatch) |
134
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
316 input_attributes = self.names2attributes(self.updateMinibatchInputAttributes()) |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
317 input_fields = minibatch(*self.updateMinibatchInputFields()) |
110
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
318 self.setAttributes(self.updateMinibatchOutputAttributes(), |
118
d0a1bd0378c6
Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
111
diff
changeset
|
319 # concatenate the attribute values and field values and then apply update fn |
134
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
320 self.update_minibatch_function(*(input_attributes+input_fields))) |
110
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
321 |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
322 def isLastEpoch(self): |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
323 """ |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
324 This method is called at the end of each epoch (cycling over the training set). |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
325 It returns a boolean to indicate if this is the last epoch. |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
326 By default just do one epoch. |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
327 """ |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
328 return True |
78 | 329 |
92
c4726e19b8ec
Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
78
diff
changeset
|
330 def update(self,training_set,train_stats_collector=None): |
107
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
331 """ |
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
332 @todo check if some of the learner attributes are actually SPECIFIED |
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
333 in as attributes of the training_set. |
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
334 """ |
110
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
335 self.updateStart(training_set) |
107
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
336 stop=False |
133 | 337 if hasattr(self,'_minibatch_size') and self._minibatch_size: |
338 minibatch_size=self._minibatch_size | |
339 else: | |
340 minibatch_size=min(100,len(training_set)) | |
107
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
341 while not stop: |
92
c4726e19b8ec
Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
78
diff
changeset
|
342 if train_stats_collector: |
107
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
343 train_stats_collector.forget() # restart stats collectin at the beginning of each epoch |
133 | 344 for minibatch in training_set.minibatches(minibatch_size=minibatch_size): |
345 self.updateMinibatch(minibatch) | |
107
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
346 if train_stats_collector: |
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
347 minibatch_set = minibatch.examples() |
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
348 minibatch_set.setAttributes(self.attributeNames(),self.attributes()) |
c4916445e025
Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
92
diff
changeset
|
349 train_stats_collector.update(minibatch_set) |
110
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
350 stop = self.isLastEpoch() |
8fa1ef2411a0
Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents:
109
diff
changeset
|
351 self.updateEnd() |
92
c4726e19b8ec
Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
78
diff
changeset
|
352 return self.use |
c4726e19b8ec
Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
78
diff
changeset
|
353 |
129
4c2280edcaf5
Fixed typos in learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
128
diff
changeset
|
354 class OnlineGradientTLearner(MinibatchUpdatesTLearner): |
111
88257dfedf8c
Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
110
diff
changeset
|
355 """ |
132
f6505ec32dc3
Updated documentation slightly
Joseph Turian <turian@gmail.com>
parents:
131
diff
changeset
|
356 Specialization of L{MinibatchUpdatesTLearner} in which the minibatch updates |
111
88257dfedf8c
Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
110
diff
changeset
|
357 are obtained by performing an online (minibatch-based) gradient step. |
88257dfedf8c
Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
110
diff
changeset
|
358 |
118
d0a1bd0378c6
Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
111
diff
changeset
|
359 Sub-classes must define the following: |
132
f6505ec32dc3
Updated documentation slightly
Joseph Turian <turian@gmail.com>
parents:
131
diff
changeset
|
360 - self._learning_rate (may be changed by the sub-class between epochs or minibatches) |
f6505ec32dc3
Updated documentation slightly
Joseph Turian <turian@gmail.com>
parents:
131
diff
changeset
|
361 - self.lossAttribute() = name of the loss field |
111
88257dfedf8c
Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
110
diff
changeset
|
362 """ |
88257dfedf8c
Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
110
diff
changeset
|
363 def __init__(self,truly_online=False): |
88257dfedf8c
Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
110
diff
changeset
|
364 """ |
88257dfedf8c
Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
110
diff
changeset
|
365 If truly_online then only one pass is made through the training set passed to update(). |
118
d0a1bd0378c6
Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
111
diff
changeset
|
366 |
d0a1bd0378c6
Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
111
diff
changeset
|
367 SUBCLASSES SHOULD CALL THIS CONSTRUCTOR ONLY AFTER HAVING DEFINED ALL THEIR THEANO FORMULAS |
111
88257dfedf8c
Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
110
diff
changeset
|
368 """ |
88257dfedf8c
Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
110
diff
changeset
|
369 self.truly_online=truly_online |
88257dfedf8c
Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
110
diff
changeset
|
370 |
118
d0a1bd0378c6
Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
111
diff
changeset
|
371 # create the formulas for the gradient update |
129
4c2280edcaf5
Fixed typos in learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
128
diff
changeset
|
372 old_params = [self.__getattribute__("_"+name) for name in self.parameterAttributes()] |
118
d0a1bd0378c6
Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
111
diff
changeset
|
373 new_params_names = ["_new_"+name for name in self.parameterAttributes()] |
129
4c2280edcaf5
Fixed typos in learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
128
diff
changeset
|
374 loss = self.__getattribute__("_"+self.lossAttribute()) |
118
d0a1bd0378c6
Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
111
diff
changeset
|
375 self.setAttributes(new_params_names, |
129
4c2280edcaf5
Fixed typos in learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
128
diff
changeset
|
376 [t.add_inplace(param,self._learning_rate*t.grad(loss,param)) |
118
d0a1bd0378c6
Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
111
diff
changeset
|
377 for param in old_params]) |
129
4c2280edcaf5
Fixed typos in learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
128
diff
changeset
|
378 MinibatchUpdatesTLearner.__init__(self) |
4c2280edcaf5
Fixed typos in learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
128
diff
changeset
|
379 |
134
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
380 |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
381 def namesOfAttributesToComputeOutputs(self,output_names): |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
382 """ |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
383 The output_names are attribute names (not the corresponding Result names, which have leading _). |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
384 Return the corresponding input names |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
385 """ |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
386 all_inputs = t.gof.graph.inputs(self.names2OpResults(output_names)) |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
387 # remove constants and leading '_' in name |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
388 |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
389 return [r.name for r in all_inputs if isinstance(r,theano.Result) and \ |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
390 not isinstance(r,theano.Constant) and not isinstance(r,theano.Value)] |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
391 #inputs = [] |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
392 #for r in all_inputs: |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
393 # if isinstance(r,theano.Result) and \ |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
394 # not isinstance(r,theano.Constant) and not isinstance(r,theano.Value): |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
395 # inputs.append(r.name) |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
396 #return inputs |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
397 |
111
88257dfedf8c
Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
110
diff
changeset
|
398 def isLastEpoch(self): |
88257dfedf8c
Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
110
diff
changeset
|
399 return self.truly_online |
88257dfedf8c
Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
110
diff
changeset
|
400 |
118
d0a1bd0378c6
Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
111
diff
changeset
|
401 def updateMinibatchInputAttributes(self): |
d0a1bd0378c6
Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
111
diff
changeset
|
402 return self.parameterAttributes() |
d0a1bd0378c6
Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
111
diff
changeset
|
403 |
d0a1bd0378c6
Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
111
diff
changeset
|
404 def updateMinibatchOutputAttributes(self): |
133 | 405 return ["new_"+name for name in self.parameterAttributes()] |
118
d0a1bd0378c6
Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
111
diff
changeset
|
406 |
d0a1bd0378c6
Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
111
diff
changeset
|
407 def updateEndInputAttributes(self): |
134
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
408 return self.namesOfAttributesToComputeOutputs(self.updateEndOutputAttributes()) |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
409 |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
410 def useInputAttributes(self): |
118
d0a1bd0378c6
Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
111
diff
changeset
|
411 return self.parameterAttributes() |
111
88257dfedf8c
Added another work in progress, for mlp's
bengioy@bengiomac.local
parents:
110
diff
changeset
|
412 |
134
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
413 def useOutputAttributes(self): |
3f4e5c9bdc5e
Fixes to ApplyFunctionDataSet and other things to make learner and mlp work
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
133
diff
changeset
|
414 return [] |
118
d0a1bd0378c6
Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
111
diff
changeset
|
415 |