Mercurial > pylearn

comparison kernel_regression.py @ 422:32c5f87bc54e

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Added __len__ to HStackedDataSet and replaced default len() by sys.maxint instead of None
author Yoshua Bengio <bengioy@iro.umontreal.ca>
date Sat, 19 Jul 2008 14:12:41 -0400
parents e01f17be270a
children e2b46a8f2b7b
comparison
equal deleted inserted replaced
421:e01f17be270a 422:32c5f87bc54e
46 posterior for f (whose mean is computed here). 46 posterior for f (whose mean is computed here).
47 47
48 48
49 Usage: 49 Usage:
50 50
51 kernel_regressor=KernelRegression(L2_regularizer=0.1,kernel=GaussianKernel(gamma=0.5)) 51 kernel_regressor=KernelRegression(L2_regularizer=0.1,gamma=0.5) (kernel=GaussianKernel(gamma=0.5))
52 kernel_predictor=kernel_regressor(training_set) 52 kernel_predictor=kernel_regressor(training_set)
53 all_results_dataset=kernel_predictor(test_set) # creates a dataset with "output" and "squared_error" field 53 all_results_dataset=kernel_predictor(test_set) # creates a dataset with "output" and "squared_error" field
54 outputs = kernel_predictor.compute_outputs(inputs) # inputs and outputs are numpy arrays 54 outputs = kernel_predictor.compute_outputs(inputs) # inputs and outputs are numpy arrays
55 outputs, errors = kernel_predictor.compute_outputs_and_errors(inputs,targets) 55 outputs, errors = kernel_predictor.compute_outputs_and_errors(inputs,targets)
56 errors = kernel_predictor.compute_errors(inputs,targets) 56 errors = kernel_predictor.compute_errors(inputs,targets)
94 M[i+1,i+1]=self.L2_regularizer 94 M[i+1,i+1]=self.L2_regularizer
95 data = trainset.fields() 95 data = trainset.fields()
96 train_inputs = numpy.array(data['input']) 96 train_inputs = numpy.array(data['input'])
97 Y[0]=1 97 Y[0]=1
98 Y[1:,:] = numpy.array(data['target']) 98 Y[1:,:] = numpy.array(data['target'])
99 M,train_inputs_square=self.equations.compute_system_matrix(train_inputs,M) 99 train_inputs_square,sumG=self.equations.compute_system_matrix(train_inputs,M)
100 M[0,1:] = sumG
101 M[1:,0] = 1
102 M[0,0] = M.shape[0]
103 print M
100 theta=numpy.linalg.solve(M,Y) 104 theta=numpy.linalg.solve(M,Y)
101 return KernelPredictor(theta,self.gamma, train_inputs, train_inputs_square) 105 return KernelPredictor(theta,self.gamma, train_inputs, train_inputs_square)
102 106
103 class KernelPredictorEquations(AutoName): 107 class KernelPredictorEquations(AutoName):
104 train_inputs = T.matrix() # n_examples x n_inputs 108 train_inputs = T.matrix() # n_examples x n_inputs
109 gamma = T.scalar() 113 gamma = T.scalar()
110 inv_gamma2 = 1./(gamma*gamma) 114 inv_gamma2 = 1./(gamma*gamma)
111 b = theta[0] 115 b = theta[0]
112 alpha = theta[1:,:] 116 alpha = theta[1:,:]
113 inputs_square = T.sum(inputs*inputs,axis=1) 117 inputs_square = T.sum(inputs*inputs,axis=1)
114 Kx = exp(-(train_inputs_square-2*dot(inputs,train_inputs.T)+inputs_square)*inv_gamma2) 118 Kx = T.exp(-(train_inputs_square-2*T.dot(inputs,train_inputs.T)+inputs_square)*inv_gamma2)
115 outputs = T.dot(Kx,alpha) + b # minibatchsize x n_outputs 119 outputs = T.dot(Kx,alpha) + b # minibatchsize x n_outputs
116 squared_errors = T.sum(T.sqr(targets-outputs),axis=1) 120 squared_errors = T.sum(T.sqr(targets-outputs),axis=1)
117 121
118 __compiled = False 122 __compiled = False
119 @classmethod 123 @classmethod
130 134
131 def __init__(self): 135 def __init__(self):
132 self.compile() 136 self.compile()
133 137
134 class KernelRegressionEquations(KernelPredictorEquations): 138 class KernelRegressionEquations(KernelPredictorEquations):
135 # P = KernelPredictorEquations
136 M = T.matrix() # (n_examples+1) x (n_examples+1) 139 M = T.matrix() # (n_examples+1) x (n_examples+1)
137 inputs = T.matrix() # n_examples x n_inputs 140 inputs = T.matrix() # n_examples x n_inputs
138 G = M[1:,1:] 141 G = M[1:,1:]
139 new_G = gemm(G,1.,inputs,inputs.T,1.) 142 new_G = T.gemm(G,1.,inputs,inputs.T,1.)
140 M2 = T.add_inplace(M,new_G) 143 sumG = T.sum(new_G,axis=0)
141 M2[0,0] = M.shape[0]
142 M2[1:,0] = 1
143 M2[0,1:] = T.sum(G,axis=0)
144 inputs_square = T.sum(inputs*inputs,axis=1) 144 inputs_square = T.sum(inputs*inputs,axis=1)
145 145
146 __compiled = False 146 __compiled = False
147 147
148 @classmethod 148 @classmethod
150 if cls.__compiled: 150 if cls.__compiled:
151 return 151 return
152 def fn(input_vars,output_vars): 152 def fn(input_vars,output_vars):
153 return staticmethod(theano.function(input_vars,output_vars, linker=linker)) 153 return staticmethod(theano.function(input_vars,output_vars, linker=linker))
154 154
155 cls.compute_system_matrix = fn([cls.inputs,cls.M],[cls.M2,cls.inputs_square]) 155 cls.compute_system_matrix = fn([cls.inputs,cls.M],[cls.inputs_square,cls.sumG])
156 156
157 cls.__compiled = True 157 cls.__compiled = True
158 158
159 def __init__(self): 159 def __init__(self):
160 self.compile() 160 self.compile()