annotate learner.py @ 203:80731832c62b

Automated merge with ssh://p-omega1@lgcm.iro.umontreal.ca/tlearn
author Frederic Bastien <bastienf@iro.umontreal.ca>
date Thu, 15 May 2008 15:21:00 -0400
parents cb6b945acf5a
children 50a8302addaf
rev   line source
1
2cd82666b9a7 Added statscollector and started writing dataset and learner.
bengioy@esprit.iro.umontreal.ca
parents: 0
diff changeset
1
172
fb4837eed1a6 fixed import of AbstractFunction
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 167
diff changeset
2 from exceptions import *
193
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
3
180
2698c0feeb54 mlp seems to work!
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 177
diff changeset
4
193
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
5 class LearningAlgorithm(object):
132
f6505ec32dc3 Updated documentation slightly
Joseph Turian <turian@gmail.com>
parents: 131
diff changeset
6 """
f6505ec32dc3 Updated documentation slightly
Joseph Turian <turian@gmail.com>
parents: 131
diff changeset
7 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
8 that allows various algorithms to be applicable to generic learning
193
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
9 algorithms. It is only given here to define the expected semantics.
1
2cd82666b9a7 Added statscollector and started writing dataset and learner.
bengioy@esprit.iro.umontreal.ca
parents: 0
diff changeset
10
132
f6505ec32dc3 Updated documentation slightly
Joseph Turian <turian@gmail.com>
parents: 131
diff changeset
11 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
12 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
13 can be applied to other data and return some output data).
193
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
14
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
15 There are two main ways of using a learning algorithms, and some learning
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
16 algorithms only support one of them. The first is the way of the standard
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
17 machine learning framework, in which a learning algorithm is applied
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
18 to a training dataset,
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
19
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
20 model = learning_algorithm(training_set)
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
21
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
22 resulting in a fully trained model that can be applied to another dataset:
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
23
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
24 output_dataset = model(input_dataset)
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
25
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
26 Note that the application of a dataset has no side-effect on the model.
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
27 In that example, the training set may for example have 'input' and 'target'
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
28 fields while the input dataset may have only 'input' (or both 'input' and
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
29 'target') and the output dataset would contain some default output fields defined
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
30 by the learning algorithm (e.g. 'output' and 'error').
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
31
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
32 The second way of using a learning algorithm is in the online or
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
33 adaptive framework, where the training data are only revealed in pieces
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
34 (maybe one example or a batch of example at a time):
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
35
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
36 model = learning_algorithm()
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
37
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
38 results in a fresh model. The model can be adapted by presenting
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
39 it with some training data,
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
40
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
41 model.update(some_training_data)
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
42 ...
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
43 model.update(some_more_training_data)
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
44 ...
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
45 model.update(yet_more_training_data)
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
46
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
47 and at any point one can use the model to perform some computation:
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
48
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
49 output_dataset = model(input_dataset)
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
50
1
2cd82666b9a7 Added statscollector and started writing dataset and learner.
bengioy@esprit.iro.umontreal.ca
parents: 0
diff changeset
51 """
193
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
52
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
53 def __init__(self): pass
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
54
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
55 def __call__(self, training_dataset=None):
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
56 """
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
57 Return a LearnerModel, either fresh (if training_dataset is None) or fully trained (otherwise).
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
58 """
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
59 raise AbstractFunction()
1
2cd82666b9a7 Added statscollector and started writing dataset and learner.
bengioy@esprit.iro.umontreal.ca
parents: 0
diff changeset
60
193
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
61 class LearnerModel(AttributesHolder):
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
62 """
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
63 LearnerModel is a base class for models returned by instances of a LearningAlgorithm subclass.
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
64 It is only given here to define the expected semantics.
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
65 """
10
80bf5492e571 Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents: 1
diff changeset
66 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
67 pass
80bf5492e571 Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents: 1
diff changeset
68
14
5ede27026e05 Working on gradient_based_learner
bengioy@bengiomac.local
parents: 13
diff changeset
69 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
70 """
80bf5492e571 Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents: 1
diff changeset
71 Continue training a learner, with the evidence provided by the given training set.
193
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
72 Hence update can be called multiple times. This is the main method used for training in the
10
80bf5492e571 Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents: 1
diff changeset
73 on-line setting or the sequential (Bayesian or not) settings.
193
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
74
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
75 This function has as side effect that self(data) will behave differently,
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
76 according to the adaptation achieved by update().
75
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents: 20
diff changeset
77
132
f6505ec32dc3 Updated documentation slightly
Joseph Turian <turian@gmail.com>
parents: 131
diff changeset
78 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
79 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
80 training.
10
80bf5492e571 Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents: 1
diff changeset
81 """
193
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
82 raise AbstractFunction()
10
80bf5492e571 Rewrote learner.py according to the specs in the wiki for learners.
bengioy@esprit.iro.umontreal.ca
parents: 1
diff changeset
83
193
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
84 def __call__(self,input_dataset,output_fieldnames=None,
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
85 test_stats_collector=None,copy_inputs=False,
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
86 put_stats_in_output_dataset=True,
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
87 output_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 """
193
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
89 A trained or partially trained L{Model} can be used with
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
90 with one or more calls to it. The argument is an input L{DataSet} (possibly
132
f6505ec32dc3 Updated documentation slightly
Joseph Turian <turian@gmail.com>
parents: 131
diff changeset
91 containing a single example) and the result is an output L{DataSet} of the same length.
128
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
92 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
93 be constructed in the output L{DataSet} (for example ['output','classification_error']).
193
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
94 Otherwise, some default output fields are produced (possibly depending on the input
cb6b945acf5a Complete redesign of learner...
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 180
diff changeset
95 fields available in the input_dataset).
20
266c68cb6136 Minor editions, plus adding untested ApplyFunctionDataset for GradientLearner in the works.
bengioy@bengiomac.local
parents: 14
diff changeset
96 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
97 visible in the output L{DataSet} returned by this method.
128
ee5507af2c60 minor edits
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 126
diff changeset
98 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
99 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
100 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
101 and attributes (which are not, e.g. 'regularization_term').
110
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
102 """
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 109
diff changeset
103 raise AbstractFunction()