annotate learner.py @ 131:57e6492644ec

Automated merge with ssh://p-omega1@lgcm.iro.umontreal.ca/tlearn
author Yoshua Bengio <bengioy@iro.umontreal.ca>
date Wed, 07 May 2008 21:40:15 -0400
parents 3d8e40e7ed18 4c2280edcaf5
children f6505ec32dc3
rev   line source
1
2cd82666b9a7 Added statscollector and started writing dataset and learner.
bengioy@esprit.iro.umontreal.ca
parents: 0
diff changeset
1
129
4c2280edcaf5 Fixed typos in learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 128
diff changeset
2 from dataset import AttributesHolder,AbstractFunction
118
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
3 import compile
129
4c2280edcaf5 Fixed typos in learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 128
diff changeset
4 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
5
110
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
6 class Learner(AttributesHolder):
1
2cd82666b9a7 Added statscollector and started writing dataset and learner.
bengioy@esprit.iro.umontreal.ca
parents: 0
diff changeset
7 """Base class for learning algorithms, provides an interface
2cd82666b9a7 Added statscollector and started writing dataset and learner.
bengioy@esprit.iro.umontreal.ca
parents: 0
diff changeset
8 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
9 algorithms.
2cd82666b9a7 Added statscollector and started writing dataset and learner.
bengioy@esprit.iro.umontreal.ca
parents: 0
diff changeset
10
10
80bf5492e571 Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents: 1
diff changeset
11 A Learner can be seen as a learning algorithm, a function that when
80bf5492e571 Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents: 1
diff changeset
12 applied to training data returns a learned function, an object that
80bf5492e571 Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents: 1
diff changeset
13 can be applied to other data and return some output data.
1
2cd82666b9a7 Added statscollector and started writing dataset and learner.
bengioy@esprit.iro.umontreal.ca
parents: 0
diff changeset
14 """
2cd82666b9a7 Added statscollector and started writing dataset and learner.
bengioy@esprit.iro.umontreal.ca
parents: 0
diff changeset
15
10
80bf5492e571 Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents: 1
diff changeset
16 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
17 pass
80bf5492e571 Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents: 1
diff changeset
18
80bf5492e571 Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents: 1
diff changeset
19 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
20 """
80bf5492e571 Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents: 1
diff changeset
21 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
22 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
23 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
24 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
25 """
80bf5492e571 Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents: 1
diff changeset
26 raise NotImplementedError
80bf5492e571 Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents: 1
diff changeset
27
14
5ede27026e05 Working on gradient_based_learner
bengioy@bengiomac.local
parents: 13
diff changeset
28 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
29 """
80bf5492e571 Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents: 1
diff changeset
30 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
31 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
32 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
33 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
34 semantics of the Learner.use method.
75
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents: 20
diff changeset
35
14
5ede27026e05 Working on gradient_based_learner
bengioy@bengiomac.local
parents: 13
diff changeset
36 The user may optionally provide a training StatsCollector that is used to record
75
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents: 20
diff changeset
37 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
38 training.
10
80bf5492e571 Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents: 1
diff changeset
39 """
13
633453635d51 Starting to work on gradient_based_learner.py
bengioy@bengiomac.local
parents: 10
diff changeset
40 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
41
80bf5492e571 Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents: 1
diff changeset
42
20
266c68cb6136 Minor editions, plus adding untested ApplyFunctionDataset for GradientLearner in the works.
bengioy@bengiomac.local
parents: 14
diff changeset
43 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
44 """
80bf5492e571 Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents: 1
diff changeset
45 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
46 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
47 """
80bf5492e571 Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents: 1
diff changeset
48 self.forget()
20
266c68cb6136 Minor editions, plus adding untested ApplyFunctionDataset for GradientLearner in the works.
bengioy@bengiomac.local
parents: 14
diff changeset
49 return self.update(learning_task,train_stats_collector)
1
2cd82666b9a7 Added statscollector and started writing dataset and learner.
bengioy@esprit.iro.umontreal.ca
parents: 0
diff changeset
50
128
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
51 def use(self,input_dataset,output_fieldnames=None,
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
52 test_stats_collector=None,copy_inputs=True,
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
53 put_stats_in_output_dataset=True,
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
54 output_attributes=[]):
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
55 """
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
56 Once a Learner has been trained by one or more call to 'update', it can
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
57 be used with one or more calls to 'use'. The argument is an input DataSet (possibly
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
58 containing a single example) and the result is an output DataSet of the same length.
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
59 If output_fieldnames is specified, it may be use to indicate which fields should
1
2cd82666b9a7 Added statscollector and started writing dataset and learner.
bengioy@esprit.iro.umontreal.ca
parents: 0
diff changeset
60 be constructed in the output DataSet (for example ['output','classification_error']).
128
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
61 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
62 Optionally, if copy_inputs, the input fields (of the input_dataset) can be made
75
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents: 20
diff changeset
63 visible in the output DataSet returned by this method.
128
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
64 Optionally, attributes of the learner can be copied in the output dataset,
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
65 and statistics computed by the stats collector also put in the output dataset.
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
66 Note the distinction between fields (which are example-wise quantities, e.g. 'input')
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
67 and attributes (which are not, e.g. 'regularization_term').
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
68
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
69 We provide here a default implementation that does all this using
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
70 a sub-class defined method: minibatchwiseUseFunction.
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
71
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
72 @todo check if some of the learner attributes are actually SPECIFIED
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
73 as attributes of the input_dataset, and if so use their values instead
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
74 of the ones in the learner.
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
75
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
76 The learner tries to compute in the output dataset the output fields specified.
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
77 If None is specified then self.defaultOutputFields(input_dataset.fieldNames())
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
78 is called to determine the output fields.
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
79
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
80 Attributes of the learner can also optionally be copied into the output dataset.
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
81 If output_attributes is None then all of the attributes in self.AttributeNames()
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
82 are copied in the output dataset, but if it is [] (the default), then none are copied.
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
83 If a test_stats_collector is provided, then its attributes (test_stats_collector.AttributeNames())
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
84 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
85 """
128
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
86 minibatchwise_use_function = self.minibatchwiseUseFunction(input_dataset.fieldNames(),
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
87 output_fieldnames,
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
88 test_stats_collector)
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
89 virtual_output_dataset = ApplyFunctionDataSet(input_dataset,
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
90 minibatchwise_use_function,
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
91 True,DataSet.numpy_vstack,
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
92 DataSet.numpy_hstack)
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
93 # actually force the computation
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
94 output_dataset = CachedDataSet(virtual_output_dataset,True)
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
95 if copy_inputs:
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
96 output_dataset = input_dataset | output_dataset
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
97 # copy the wanted attributes in the dataset
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
98 if output_attributes is None:
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
99 output_attributes = self.attributeNames()
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
100 if output_attributes:
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
101 assert set(attribute_names) <= set(self.attributeNames())
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
102 output_dataset.setAttributes(output_attributes,
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
103 self.names2attributes(output_attributes,return_copy=True))
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
104 if test_stats_collector:
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
105 test_stats_collector.update(output_dataset)
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
106 if put_stats_in_output_dataset:
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
107 output_dataset.setAttributes(test_stats_collector.attributeNames(),
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
108 test_stats_collector.attributes())
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
109 return output_dataset
77
1e2bb5bad636 toying with different ways to implement learners
bengioy@bengiomac.local
parents: 75
diff changeset
110
128
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
111 def minibatchwiseUseFunction(self, input_fields, output_fields, stats_collector):
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
112 """
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
113 Returns a function that can map the given input fields to the given output fields
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
114 and to the attributes that the stats collector needs for its computation.
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
115 That function is expected to operate on minibatches.
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
116 The function returned makes use of the self.useInputAttributes() and
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
117 sets the attributes specified by self.useOutputAttributes().
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
118 """
92
c4726e19b8ec Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 78
diff changeset
119 def attributeNames(self):
78
3499918faa9d In the middle of designing TLearner
bengioy@bengiomac.local
parents: 77
diff changeset
120 """
3499918faa9d In the middle of designing TLearner
bengioy@bengiomac.local
parents: 77
diff changeset
121 A Learner may have attributes that it wishes to export to other objects. To automate
3499918faa9d In the middle of designing TLearner
bengioy@bengiomac.local
parents: 77
diff changeset
122 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
123
c4916445e025 Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 92
diff changeset
124 @todo By default, attributeNames looks for all dictionary entries whose name does not start with _.
78
3499918faa9d In the middle of designing TLearner
bengioy@bengiomac.local
parents: 77
diff changeset
125 """
3499918faa9d In the middle of designing TLearner
bengioy@bengiomac.local
parents: 77
diff changeset
126 return []
77
1e2bb5bad636 toying with different ways to implement learners
bengioy@bengiomac.local
parents: 75
diff changeset
127
128
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
128 def attributes(self,return_copy=False):
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
129 """
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
130 Return a list with the values of the learner's attributes (or optionally, a deep copy).
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
131 """
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
132 return self.names2attributes(self.attributeNames(),return_copy)
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
133
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
134 def names2attributes(self,names,return_copy=False):
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
135 """
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
136 Private helper function that maps a list of attribute names to a list
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
137 of (optionally copies) values of attributes.
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
138 """
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
139 if return_copy:
129
4c2280edcaf5 Fixed typos in learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 128
diff changeset
140 return [copy.deepcopy(self.__getattribute__(name).data) for name in names]
128
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
141 else:
129
4c2280edcaf5 Fixed typos in learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 128
diff changeset
142 return [self.__getattribute__(name).data for name in names]
128
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
143
110
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
144 def updateInputAttributes(self):
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
145 """
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
146 A subset of self.attributeNames() which are the names of attributes needed by update() in order
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
147 to do its work.
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
148 """
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
149 raise AbstractFunction()
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
150
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
151 def useInputAttributes(self):
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
152 """
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
153 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
154 to do its work.
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
155 """
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
156 raise AbstractFunction()
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
157
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
158 def updateOutputAttributes(self):
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 A subset of self.attributeNames() which are the names of attributes modified/created by update() in order
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
161 to do its work.
111
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents: 110
diff changeset
162
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents: 110
diff changeset
163 By default these are inferred from the various update output attributes:
110
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
164 """
111
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents: 110
diff changeset
165 return ["parameters"] + self.updateMinibatchOutputAttributes() + self.updateEndOutputAttributes()
110
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 def useOutputAttributes(self):
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 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
170 to do its work.
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
171 """
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
172 raise AbstractFunction()
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
173
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
174
78
3499918faa9d In the middle of designing TLearner
bengioy@bengiomac.local
parents: 77
diff changeset
175 class TLearner(Learner):
3499918faa9d In the middle of designing TLearner
bengioy@bengiomac.local
parents: 77
diff changeset
176 """
3499918faa9d In the middle of designing TLearner
bengioy@bengiomac.local
parents: 77
diff changeset
177 TLearner is a virtual class of Learners that attempts to factor out of the definition
3499918faa9d In the middle of designing TLearner
bengioy@bengiomac.local
parents: 77
diff changeset
178 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
179 so as to leave only 'the equations' to define in particular sub-classes, using Theano.
78
3499918faa9d In the middle of designing TLearner
bengioy@bengiomac.local
parents: 77
diff changeset
180
3499918faa9d In the middle of designing TLearner
bengioy@bengiomac.local
parents: 77
diff changeset
181 In the default implementations of use and update, it is assumed that the 'use' and 'update' methods
3499918faa9d In the middle of designing TLearner
bengioy@bengiomac.local
parents: 77
diff changeset
182 visit examples in the input dataset sequentially. In the 'use' method only one pass through the dataset is done,
3499918faa9d In the middle of designing TLearner
bengioy@bengiomac.local
parents: 77
diff changeset
183 whereas the sub-learner may wish to iterate over the examples multiple times. Subclasses where this
3499918faa9d In the middle of designing TLearner
bengioy@bengiomac.local
parents: 77
diff changeset
184 basic model is not appropriate can simply redefine update or use.
3499918faa9d In the middle of designing TLearner
bengioy@bengiomac.local
parents: 77
diff changeset
185
3499918faa9d In the middle of designing TLearner
bengioy@bengiomac.local
parents: 77
diff changeset
186 Sub-classes must provide the following functions and functionalities:
3499918faa9d In the middle of designing TLearner
bengioy@bengiomac.local
parents: 77
diff changeset
187 - attributeNames(): defines all the names of attributes which can be used as fields or
3499918faa9d In the middle of designing TLearner
bengioy@bengiomac.local
parents: 77
diff changeset
188 attributes in input/output datasets or in stats collectors.
3499918faa9d In the middle of designing TLearner
bengioy@bengiomac.local
parents: 77
diff changeset
189 All these attributes are expected to be theano.Result objects
3499918faa9d In the middle of designing TLearner
bengioy@bengiomac.local
parents: 77
diff changeset
190 (with a .data property and recognized by theano.Function for compilation).
3499918faa9d In the middle of designing TLearner
bengioy@bengiomac.local
parents: 77
diff changeset
191 The sub-class constructor defines the relations between
3499918faa9d In the middle of designing TLearner
bengioy@bengiomac.local
parents: 77
diff changeset
192 the Theano variables that may be used by 'use' and 'update'
3499918faa9d In the middle of designing TLearner
bengioy@bengiomac.local
parents: 77
diff changeset
193 or by a stats collector.
3499918faa9d In the middle of designing TLearner
bengioy@bengiomac.local
parents: 77
diff changeset
194 - defaultOutputFields(input_fields): return a list of default dataset output fields when
3499918faa9d In the middle of designing TLearner
bengioy@bengiomac.local
parents: 77
diff changeset
195 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
196 The following naming convention is assumed and important.
c4726e19b8ec Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 78
diff changeset
197 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
198 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
199 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
200 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
201 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
202 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
203 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
204 Theano Tensor (for symbolic manipulation).
107
c4916445e025 Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 92
diff changeset
205
c4916445e025 Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 92
diff changeset
206 @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
207 dependant de Theano
78
3499918faa9d In the middle of designing TLearner
bengioy@bengiomac.local
parents: 77
diff changeset
208 """
92
c4726e19b8ec Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 78
diff changeset
209
c4726e19b8ec Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 78
diff changeset
210 def __init__(self):
c4726e19b8ec Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 78
diff changeset
211 Learner.__init__(self)
110
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
212
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
213 def defaultOutputFields(self, input_fields):
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
214 """
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
215 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
216 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
217 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
218 input_dataset.
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
219 """
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
220 raise AbstractFunction()
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
221
128
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
222 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
223 """
128
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
224 Implement minibatchwiseUseFunction by exploiting Theano compilation
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
225 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
226 """
c4726e19b8ec Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 78
diff changeset
227 if not output_fields:
c4726e19b8ec Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 78
diff changeset
228 output_fields = self.defaultOutputFields(input_fields)
c4726e19b8ec Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 78
diff changeset
229 if stats_collector:
110
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
230 stats_collector_inputs = stats_collector.input2UpdateAttributes()
92
c4726e19b8ec Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 78
diff changeset
231 for attribute in stats_collector_inputs:
c4726e19b8ec Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 78
diff changeset
232 if attribute not in input_fields:
c4726e19b8ec Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 78
diff changeset
233 output_fields.append(attribute)
c4726e19b8ec Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 78
diff changeset
234 key = (input_fields,output_fields)
c4726e19b8ec Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 78
diff changeset
235 if key not in self.use_functions_dictionary:
110
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
236 use_input_attributes = self.useInputAttributes()
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
237 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
238 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
239 self.names2OpResults(output_fields+use_output_attributes))
110
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
240 def f(*input_field_values):
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
241 input_attribute_values = self.names2attributes(use_input_attributes)
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
242 results = complete_f(*(input_field_values + input_attribute_values))
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
243 output_field_values = results[0:len(output_fields)]
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
244 output_attribute_values = results[len(output_fields):len(results)]
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
245 if use_output_attributes:
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
246 self.setAttributes(use_output_attributes,output_attribute_values)
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
247 return output_field_values
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
248 self.use_functions_dictionary[key]=f
92
c4726e19b8ec Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 78
diff changeset
249 return self.use_functions_dictionary[key]
c4726e19b8ec Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 78
diff changeset
250
110
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
251 def names2OpResults(self,names):
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
252 """
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
253 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
254 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
255 """
129
4c2280edcaf5 Fixed typos in learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 128
diff changeset
256 return [self.__getattribute__('_'+name).data for name in names]
92
c4726e19b8ec Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 78
diff changeset
257
107
c4916445e025 Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 92
diff changeset
258
111
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents: 110
diff changeset
259 class MinibatchUpdatesTLearner(TLearner):
107
c4916445e025 Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 92
diff changeset
260 """
c4916445e025 Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 92
diff changeset
261 This adds to TLearner a
110
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
262 - updateStart(), updateEnd(), updateMinibatch(minibatch), isLastEpoch():
107
c4916445e025 Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 92
diff changeset
263 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
264 (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
265 of each epoch. This model only
c4916445e025 Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 92
diff changeset
266 works for 'online' or one-shot learning that requires
c4916445e025 Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 92
diff changeset
267 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
268 models, more specialized subclasses of TLearner should be used
c4916445e025 Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 92
diff changeset
269 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
270
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents: 110
diff changeset
271 - 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
272 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
273
107
c4916445e025 Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 92
diff changeset
274 """
c4916445e025 Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 92
diff changeset
275
c4916445e025 Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 92
diff changeset
276 def __init__(self):
c4916445e025 Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 92
diff changeset
277 TLearner.__init__(self)
118
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
278 self.update_minibatch_function = compile.function
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
279 (self.names2OpResults(self.updateMinibatchOutputAttributes()+
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
280 self.updateMinibatchInputFields()),
110
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
281 self.names2OpResults(self.updateMinibatchOutputAttributes()))
118
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
282 self.update_end_function = compile.function
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
283 (self.names2OpResults(self.updateEndInputAttributes()),
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
284 self.names2OpResults(self.updateEndOutputAttributes()))
110
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
285
128
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
286 def allocate(self, minibatch):
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
287 """
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
288 This function is called at the beginning of each updateMinibatch
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
289 and should be used to check that all required attributes have been
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
290 allocated and initialized (usually this function calls forget()
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
291 when it has to do an initialization).
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
292 """
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
293 raise AbstractFunction()
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
294
110
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
295 def updateMinibatchInputFields(self):
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
296 raise AbstractFunction()
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
297
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
298 def updateMinibatchInputAttributes(self):
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
299 raise AbstractFunction()
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
300
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
301 def updateMinibatchOutputAttributes(self):
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
302 raise AbstractFunction()
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
303
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
304 def updateEndInputAttributes(self):
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
305 raise AbstractFunction()
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
306
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
307 def updateEndOutputAttributes(self):
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
308 raise AbstractFunction()
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
309
111
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents: 110
diff changeset
310 def parameterAttributes(self):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents: 110
diff changeset
311 raise AbstractFunction()
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents: 110
diff changeset
312
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents: 110
diff changeset
313 def updateStart(self): pass
110
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
314
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
315 def updateEnd(self):
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
316 self.setAttributes(self.updateEndOutputAttributes(),
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
317 self.update_end_function
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
318 (self.names2attributes(self.updateEndInputAttributes())))
111
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents: 110
diff changeset
319 self.parameters = self.names2attributes(self.parameterAttributes())
107
c4916445e025 Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 92
diff changeset
320
110
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
321 def updateMinibatch(self,minibatch):
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
322 # make sure all required fields are allocated and initialized
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
323 self.allocate(minibatch)
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
324 self.setAttributes(self.updateMinibatchOutputAttributes(),
118
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
325 # concatenate the attribute values and field values and then apply update fn
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
326 self.update_minibatch_function(*(self.names2attributes
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
327 (self.updateMinibatchInputAttributes()))
110
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
328 + minibatch(self.updateMinibatchInputFields())))
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
329
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
330 def isLastEpoch(self):
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
331 """
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
332 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
333 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
334 By default just do one epoch.
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
335 """
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
336 return True
78
3499918faa9d In the middle of designing TLearner
bengioy@bengiomac.local
parents: 77
diff changeset
337
92
c4726e19b8ec Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 78
diff changeset
338 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
339 """
c4916445e025 Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 92
diff changeset
340 @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
341 in as attributes of the training_set.
c4916445e025 Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 92
diff changeset
342 """
110
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
343 self.updateStart(training_set)
107
c4916445e025 Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 92
diff changeset
344 stop=False
c4916445e025 Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 92
diff changeset
345 while not stop:
92
c4726e19b8ec Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 78
diff changeset
346 if train_stats_collector:
107
c4916445e025 Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 92
diff changeset
347 train_stats_collector.forget() # restart stats collectin at the beginning of each epoch
c4916445e025 Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 92
diff changeset
348 for minibatch in training_set.minibatches(self.training_set_input_fields,
c4916445e025 Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 92
diff changeset
349 minibatch_size=self.minibatch_size):
c4916445e025 Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 92
diff changeset
350 self.update_minibatch(minibatch)
c4916445e025 Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 92
diff changeset
351 if train_stats_collector:
c4916445e025 Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 92
diff changeset
352 minibatch_set = minibatch.examples()
c4916445e025 Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 92
diff changeset
353 minibatch_set.setAttributes(self.attributeNames(),self.attributes())
c4916445e025 Comments from Pascal V.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 92
diff changeset
354 train_stats_collector.update(minibatch_set)
110
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
355 stop = self.isLastEpoch()
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
356 self.updateEnd()
92
c4726e19b8ec Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 78
diff changeset
357 return self.use
c4726e19b8ec Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 78
diff changeset
358
129
4c2280edcaf5 Fixed typos in learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 128
diff changeset
359 class OnlineGradientTLearner(MinibatchUpdatesTLearner):
111
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents: 110
diff changeset
360 """
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents: 110
diff changeset
361 Specialization of MinibatchUpdatesTLearner in which the minibatch updates
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents: 110
diff changeset
362 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
363
118
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
364 Sub-classes must define the following:
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
365
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
366 self._learning_rate (may be changed by the sub-class between epochs or minibatches)
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
367
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
368 self.lossAttribute() = name of the loss field
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
369
111
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents: 110
diff changeset
370 """
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents: 110
diff changeset
371 def __init__(self,truly_online=False):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents: 110
diff changeset
372 """
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents: 110
diff changeset
373 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
374
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
375 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
376 """
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents: 110
diff changeset
377 self.truly_online=truly_online
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents: 110
diff changeset
378
118
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
379 # create the formulas for the gradient update
129
4c2280edcaf5 Fixed typos in learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 128
diff changeset
380 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
381 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
382 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
383 self.setAttributes(new_params_names,
129
4c2280edcaf5 Fixed typos in learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 128
diff changeset
384 [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
385 for param in old_params])
129
4c2280edcaf5 Fixed typos in learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 128
diff changeset
386 MinibatchUpdatesTLearner.__init__(self)
4c2280edcaf5 Fixed typos in learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 128
diff changeset
387
111
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents: 110
diff changeset
388 def isLastEpoch(self):
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents: 110
diff changeset
389 return self.truly_online
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents: 110
diff changeset
390
118
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
391 def updateMinibatchInputAttributes(self):
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
392 return self.parameterAttributes()
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
393
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
394 def updateMinibatchOutputAttributes(self):
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
395 return ["_new"+name for name in self.parameterAttributes()]
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
396
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
397 def updateEndInputAttributes(self):
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
398 return self.parameterAttributes()
111
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents: 110
diff changeset
399
118
d0a1bd0378c6 Finished draft of OneHiddenLayerNNetClassifier to debut learner.py
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 111
diff changeset
400