Mercurial > pylearn

comparison gradient_learner.py @ 23:526e192b0699

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Working on ApplyFunctionDataSet, added constraint that DataSet iterators must have a next_index() method.
author bengioy@esprit.iro.umontreal.ca
date Wed, 09 Apr 2008 18:27:13 -0400
parents 266c68cb6136
children 672fe4b23032
comparison
equal deleted inserted replaced
22:b6b36f65664f 23:526e192b0699
7 class GradientLearner(Learner): 7 class GradientLearner(Learner):
8 """ 8 """
9 Base class for gradient-based optimization of a training criterion 9 Base class for gradient-based optimization of a training criterion
10 that can consist in two parts, an additive part over examples, and 10 that can consist in two parts, an additive part over examples, and
11 an example-independent part (usually called the regularizer). 11 an example-independent part (usually called the regularizer).
12 The user provides a Theano formula that maps the fields of a training example 12 The user provides a Theano formula that maps the fields of a minibatch (each being a tensor with the
13 and parameters to output fields (for the use function), one of which must be a cost 13 same number of rows = minibatch size) and parameters to output fields (for the use function), one of which
14 that is the training criterion to be minimized. Subclasses implement 14 must be a cost that is the training criterion to be minimized. Subclasses implement
15 a training strategy that uses the Theano formula to compute gradients and 15 a training strategy that uses the Theano formula to compute gradients and
16 to compute outputs in the update method. 16 to compute outputs in the update method.
17 The inputs, parameters, and outputs are lists of Theano tensors, 17 The inputs, parameters, and outputs are lists of Theano tensors,
18 while the example_wise_cost and regularization_term are Theano tensors. 18 while the example_wise_cost and regularization_term are Theano tensors.
19 The user can specify a regularization coefficient that multiplies the regularization term. 19 The user can specify a regularization coefficient that multiplies the regularization term.
53 53
54 use_function_key = input_fields+output_fields 54 use_function_key = input_fields+output_fields
55 if not self.use_functions.has_key(use_function_key): 55 if not self.use_functions.has_key(use_function_key):
56 self.use_function[use_function_key]=Function(input_fields,output_fields) 56 self.use_function[use_function_key]=Function(input_fields,output_fields)
57 use_function = self.use_functions[use_function_key] 57 use_function = self.use_functions[use_function_key]
58 # return a virtual dataset that computes the outputs on demand 58 # return a dataset that computes the outputs
59 return input_dataset.apply_function(use_function,input_fields,output_fields,copy_inputs,accept_minibatches=???) 59 return input_dataset.applyFunction(use_function,input_fields,output_fields,copy_inputs,compute_now=True)
60 60