Mercurial > pylearn

annotate pylearn/algorithms/linear_regression.py @ 1505:723e2d761985

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
auto white space fix.
author Frederic Bastien <nouiz@nouiz.org>
date Mon, 12 Sep 2011 10:49:15 -0400
parents bf5c0f797161
children
rev   line source
132
f6505ec32dc3 Updated documentation slightly
Joseph Turian <turian@gmail.com>
parents: 118
diff changeset
1 """
f6505ec32dc3 Updated documentation slightly
Joseph Turian <turian@gmail.com>
parents: 118
diff changeset
2 Implementation of linear regression, with or without L2 regularization.
f6505ec32dc3 Updated documentation slightly
Joseph Turian <turian@gmail.com>
parents: 118
diff changeset
3 This is one of the simplest example of L{learner}, and illustrates
f6505ec32dc3 Updated documentation slightly
Joseph Turian <turian@gmail.com>
parents: 118
diff changeset
4 the use of theano.
f6505ec32dc3 Updated documentation slightly
Joseph Turian <turian@gmail.com>
parents: 118
diff changeset
5 """
75
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
6
1504
bf5c0f797161 Fix test.
Frederic Bastien <nouiz@nouiz.org>
parents: 732
diff changeset
7 #from pylearn.old_dataset.learner import OfflineLearningAlgorithm,OnlineLearningAlgorithm
376
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
8 from theano import tensor as T
523
111e547ffa7b modified to use the new implecement of ops and use the new interface to theano.function
Frederic Bastien <bastienf@iro.umontreal.ca>
parents: 432
diff changeset
9 from theano.tensor.nnet import prepend_1_to_each_row
75
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
10 from theano.scalar import as_scalar
376
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
11 from common.autoname import AutoName
390
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
12 import theano
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
13 import numpy
75
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
14
1504
bf5c0f797161 Fix test.
Frederic Bastien <nouiz@nouiz.org>
parents: 732
diff changeset
15 class LinearRegression():#OfflineLearningAlgorithm):
75
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
16 """
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
17 Implement linear regression, with or without L2 regularization
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
18 (the former is called Ridge Regression and the latter Ordinary Least Squares).
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
19
390
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
20 Usage:
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
21
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
22 linear_regressor=LinearRegression(L2_regularizer=0.1)
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
23 linear_predictor=linear_regression(training_set)
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
24 all_results_dataset=linear_predictor(test_set) # creates a dataset with "output" and "squared_error" field
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
25 outputs = linear_predictor.compute_outputs(inputs) # inputs and outputs are numpy arrays
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
26 outputs, errors = linear_predictor.compute_outputs_and_errors(inputs,targets)
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
27 errors = linear_predictor.compute_errors(inputs,targets)
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
28 mse = linear_predictor.compute_mse(inputs,targets)
1505
723e2d761985 auto white space fix.
Frederic Bastien <nouiz@nouiz.org>
parents: 1504
diff changeset
29
723e2d761985 auto white space fix.
Frederic Bastien <nouiz@nouiz.org>
parents: 1504
diff changeset
30
390
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
31
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
32 The training_set must have fields "input" and "target".
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
33 The test_set must have field "input", and needs "target" if
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
34 we want to compute the squared errors.
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
35
92
c4726e19b8ec Finished first draft of TLearner
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 78
diff changeset
36 The predictor parameters are obtained analytically from the training set.
421
e01f17be270a Kernel regression learning algorithm
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 419
diff changeset
37
75
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
38 For each (input[t],output[t]) pair in a minibatch,::
1505
723e2d761985 auto white space fix.
Frederic Bastien <nouiz@nouiz.org>
parents: 1504
diff changeset
39
75
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
40 output_t = b + W * input_t
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
41
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
42 where b and W are obtained by minimizing::
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
43
111
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents: 110
diff changeset
44 L2_regularizer sum_{ij} W_{ij}^2 + sum_t ||output_t - target_t||^2
75
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
45
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
46 Let X be the whole training set inputs matrix (one input example per row),
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
47 with the first column full of 1's, and Let Y the whole training set
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
48 targets matrix (one example's target vector per row).
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
49 Let theta = the matrix with b in its first column and W in the others,
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
50 then each theta[:,i] is the solution of the linear system::
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
51
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
52 XtX * theta[:,i] = XtY[:,i]
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
53
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
54 where XtX is a (n_inputs+1)x(n_inputs+1) matrix containing X'*X
111
88257dfedf8c Added another work in progress, for mlp's
bengioy@bengiomac.local
parents: 110
diff changeset
55 plus L2_regularizer on the diagonal except at (0,0),
75
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
56 and XtY is a (n_inputs+1)*n_outputs matrix containing X'*Y.
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
57
376
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
58 The dataset fields expected and produced by the learning algorithm and the trained model
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
59 are the following:
75
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
60
376
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
61 - Input and output dataset fields (example-wise quantities):
75
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
62
376
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
63 - 'input' (always expected as an input_dataset field)
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
64 - 'target' (always expected by the learning algorithm, optional for learned model)
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
65 - 'output' (always produced by learned model)
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
66 - 'squared_error' (optionally produced by learned model if 'target' is provided)
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
67 = example-wise squared error
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
68 """
390
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
69 def __init__(self, L2_regularizer=0,minibatch_size=10000):
376
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
70 self.L2_regularizer=L2_regularizer
390
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
71 self.equations = LinearRegressionEquations()
421
e01f17be270a Kernel regression learning algorithm
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 419
diff changeset
72 self.minibatch_size=minibatch_size
390
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
73
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
74 def __call__(self,trainset):
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
75 first_example = trainset[0]
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
76 n_inputs = first_example['input'].size
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
77 n_outputs = first_example['target'].size
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
78 XtX = numpy.zeros((n_inputs+1,n_inputs+1))
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
79 XtY = numpy.zeros((n_inputs+1,n_outputs))
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
80 for i in xrange(n_inputs):
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
81 XtX[i+1,i+1]=self.L2_regularizer
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
82 mbs=min(self.minibatch_size,len(trainset))
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
83 for inputs,targets in trainset.minibatches(["input","target"],minibatch_size=mbs):
403
273e5c03003e Making linear_regression more robust
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 390
diff changeset
84 XtX,XtY=self.equations.update(XtX,XtY,numpy.array(inputs),numpy.array(targets))
390
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
85 theta=numpy.linalg.solve(XtX,XtY)
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
86 return LinearPredictor(theta)
75
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
87
376
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
88 class LinearPredictorEquations(AutoName):
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
89 inputs = T.matrix() # minibatchsize x n_inputs
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
90 targets = T.matrix() # minibatchsize x n_outputs
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
91 theta = T.matrix() # (n_inputs+1) x n_outputs
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
92 b = theta[0]
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
93 Wt = theta[1:,:]
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
94 outputs = T.dot(inputs,Wt) + b # minibatchsize x n_outputs
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
95 squared_errors = T.sum(T.sqr(targets-outputs),axis=1)
75
90e4c0784d6e Added draft of LinearRegression learner
bengioy@bengiomac.local
parents:
diff changeset
96
376
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
97 __compiled = False
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
98 @classmethod
523
111e547ffa7b modified to use the new implecement of ops and use the new interface to theano.function
Frederic Bastien <bastienf@iro.umontreal.ca>
parents: 432
diff changeset
99 def compile(cls, mode = "FAST_RUN"):
376
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
100 if cls.__compiled:
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
101 return
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
102 def fn(input_vars,output_vars):
523
111e547ffa7b modified to use the new implecement of ops and use the new interface to theano.function
Frederic Bastien <bastienf@iro.umontreal.ca>
parents: 432
diff changeset
103 return staticmethod(theano.function(input_vars, output_vars, mode=mode))
110
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 107
diff changeset
104
528
cfe3f62a08cb bugfix, outputs needed to be in a list in the past, not anymore.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents: 523
diff changeset
105 cls.compute_outputs = fn([cls.inputs,cls.theta],cls.outputs)
cfe3f62a08cb bugfix, outputs needed to be in a list in the past, not anymore.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents: 523
diff changeset
106 cls.compute_errors = fn([cls.outputs,cls.targets],cls.squared_errors)
110
8fa1ef2411a0 Worked on OneShotTLearner and implementation of LinearRegression
bengioy@bengiomac.local
parents: 107
diff changeset
107
376
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
108 cls.__compiled = True
77
1e2bb5bad636 toying with different ways to implement learners
bengioy@bengiomac.local
parents: 75
diff changeset
109
390
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
110 def __init__(self):
376
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
111 self.compile()
1505
723e2d761985 auto white space fix.
Frederic Bastien <nouiz@nouiz.org>
parents: 1504
diff changeset
112
376
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
113 class LinearRegressionEquations(LinearPredictorEquations):
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
114 P = LinearPredictorEquations
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
115 XtX = T.matrix() # (n_inputs+1) x (n_inputs+1)
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
116 XtY = T.matrix() # (n_inputs+1) x n_outputs
390
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
117 extended_input = prepend_1_to_each_row(P.inputs)
523
111e547ffa7b modified to use the new implecement of ops and use the new interface to theano.function
Frederic Bastien <bastienf@iro.umontreal.ca>
parents: 432
diff changeset
118 new_XtX = T.add(XtX,T.dot(extended_input.T,extended_input))
111e547ffa7b modified to use the new implecement of ops and use the new interface to theano.function
Frederic Bastien <bastienf@iro.umontreal.ca>
parents: 432
diff changeset
119 new_XtY = T.add(XtY,T.dot(extended_input.T,P.targets))
390
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
120
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
121 __compiled = False
1505
723e2d761985 auto white space fix.
Frederic Bastien <nouiz@nouiz.org>
parents: 1504
diff changeset
122
390
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
123 @classmethod
523
111e547ffa7b modified to use the new implecement of ops and use the new interface to theano.function
Frederic Bastien <bastienf@iro.umontreal.ca>
parents: 432
diff changeset
124 def compile(cls, mode="FAST_RUN"):
390
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
125 if cls.__compiled:
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
126 return
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
127 def fn(input_vars,output_vars):
523
111e547ffa7b modified to use the new implecement of ops and use the new interface to theano.function
Frederic Bastien <bastienf@iro.umontreal.ca>
parents: 432
diff changeset
128 return staticmethod(theano.function(input_vars, output_vars, mode=mode))
390
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
129
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
130 cls.update = fn([cls.XtX,cls.XtY,cls.P.inputs,cls.P.targets],[cls.new_XtX,cls.new_XtY])
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
131
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
132 cls.__compiled = True
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
133
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
134 def __init__(self):
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
135 self.compile()
379
74b402b5a81b small modif by yoshue
Frederic Bastien <bastienf@iro.umontreal.ca>
parents: 376
diff changeset
136
376
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
137 class LinearPredictor(object):
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
138 """
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
139 A linear predictor has parameters theta (a bias vector and a weight matrix)
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
140 it can use to make a linear prediction (according to the LinearPredictorEquations).
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
141 It can compute its output (bias + weight * input) and a squared error (||output - target||^2).
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
142 """
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
143 def __init__(self, theta):
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
144 self.theta=theta
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
145 self.n_inputs=theta.shape[0]-1
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
146 self.n_outputs=theta.shape[1]
390
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
147 self.equations = LinearPredictorEquations()
77
1e2bb5bad636 toying with different ways to implement learners
bengioy@bengiomac.local
parents: 75
diff changeset
148
376
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
149 def compute_outputs(self,inputs):
390
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
150 return self.equations.compute_outputs(inputs,self.theta)
376
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
151 def compute_errors(self,inputs,targets):
390
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
152 return self.equations.compute_errors(self.compute_outputs(inputs),targets)
376
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
153 def compute_outputs_and_errors(self,inputs,targets):
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
154 outputs = self.compute_outputs(inputs)
390
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
155 return [outputs,self.equations.compute_errors(outputs,targets)]
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
156 def compute_mse(self,inputs,targets):
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
157 errors = self.compute_errors(inputs,targets)
efb797c5efc0 First non-crashing draft of LinearRegression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 379
diff changeset
158 return numpy.sum(errors)/errors.size
1505
723e2d761985 auto white space fix.
Frederic Bastien <nouiz@nouiz.org>
parents: 1504
diff changeset
159
376
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
160 def __call__(self,dataset,output_fieldnames=None,cached_output_dataset=False):
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
161 assert dataset.hasFields(["input"])
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
162 if output_fieldnames is None:
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
163 if dataset.hasFields(["target"]):
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
164 output_fieldnames = ["output","squared_error"]
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
165 else:
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
166 output_fieldnames = ["output"]
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
167 output_fieldnames.sort()
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
168 if output_fieldnames == ["squared_error"]:
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
169 f = self.compute_errors
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
170 elif output_fieldnames == ["output"]:
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
171 f = self.compute_outputs
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
172 elif output_fieldnames == ["output","squared_error"]:
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
173 f = self.compute_outputs_and_errors
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
174 else:
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
175 raise ValueError("unknown field(s) in output_fieldnames: "+str(output_fieldnames))
1505
723e2d761985 auto white space fix.
Frederic Bastien <nouiz@nouiz.org>
parents: 1504
diff changeset
176
376
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
177 ds=ApplyFunctionDataSet(dataset,f,output_fieldnames)
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
178 if cached_output_dataset:
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
179 return CachedDataSet(ds)
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
180 else:
c9a89be5cb0a Redesigning linear_regression
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 132
diff changeset
181 return ds
1505
723e2d761985 auto white space fix.
Frederic Bastien <nouiz@nouiz.org>
parents: 1504
diff changeset
182
77
1e2bb5bad636 toying with different ways to implement learners
bengioy@bengiomac.local
parents: 75
diff changeset
183
431
0f8c81b0776d Adding file make_test_datasets to host simple data-generating processes
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 427
diff changeset
184 def linear_predictor(inputs,params,*otherargs):
1505
723e2d761985 auto white space fix.
Frederic Bastien <nouiz@nouiz.org>
parents: 1504
diff changeset
185 p = LinearPredictor(params)
723e2d761985 auto white space fix.
Frederic Bastien <nouiz@nouiz.org>
parents: 1504
diff changeset
186 return p.compute_outputs(inputs)
431
0f8c81b0776d Adding file make_test_datasets to host simple data-generating processes
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 427
diff changeset
187
427
fa4a5fee53ce Showing the path to an online version of linear regressor.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 421
diff changeset
188 #TODO : an online version
1504
bf5c0f797161 Fix test.
Frederic Bastien <nouiz@nouiz.org>
parents: 732
diff changeset
189 class OnlineLinearRegression():#OnlineLearningAlgorithm):
427
fa4a5fee53ce Showing the path to an online version of linear regressor.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 421
diff changeset
190 """
fa4a5fee53ce Showing the path to an online version of linear regressor.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 421
diff changeset
191 Training can proceed sequentially (with multiple calls to update with
fa4a5fee53ce Showing the path to an online version of linear regressor.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 421
diff changeset
192 different disjoint subsets of the training sets). After each call to
fa4a5fee53ce Showing the path to an online version of linear regressor.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 421
diff changeset
193 update the predictor is ready to be used (and optimized for the union
fa4a5fee53ce Showing the path to an online version of linear regressor.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 421
diff changeset
194 of all the training sets passed to update since construction or since
fa4a5fee53ce Showing the path to an online version of linear regressor.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 421
diff changeset
195 the last call to forget).
fa4a5fee53ce Showing the path to an online version of linear regressor.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 421
diff changeset
196 """
fa4a5fee53ce Showing the path to an online version of linear regressor.
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 421
diff changeset
197 pass