Mercurial > pylearn
comparison learner.py @ 167:4803cb76e26b
Updated documentation
| author | Joseph Turian <turian@gmail.com> |
|---|---|
| date | Mon, 12 May 2008 18:51:42 -0400 |
| parents | ceae4de18981 |
| children | fb4837eed1a6 |
comparison
equal
deleted
inserted
replaced
| 166:ee11ed427ba8 | 167:4803cb76e26b |
|---|---|
| 11 algorithms. | 11 algorithms. |
| 12 | 12 |
| 13 A L{Learner} can be seen as a learning algorithm, a function that when | 13 A L{Learner} can be seen as a learning algorithm, a function that when |
| 14 applied to training data returns a learned function (which is an object that | 14 applied to training data returns a learned function (which is an object that |
| 15 can be applied to other data and return some output data). | 15 can be applied to other data and return some output data). |
| 16 | |
| 17 """ | 16 """ |
| 18 | 17 |
| 19 def __init__(self): | 18 def __init__(self): |
| 20 pass | 19 pass |
| 21 | 20 |
| 167 raise AbstractFunction() | 166 raise AbstractFunction() |
| 168 | 167 |
| 169 | 168 |
| 170 class TLearner(Learner): | 169 class TLearner(Learner): |
| 171 """ | 170 """ |
| 172 TLearner is a virtual class of Learners that attempts to factor out of the definition | 171 TLearner is a virtual class of L{Learner}s that attempts to factor |
| 173 of a learner the steps that are common to many implementations of learning algorithms, | 172 out of the definition of a learner the steps that are common to |
| 174 so as to leave only 'the equations' to define in particular sub-classes, using Theano. | 173 many implementations of learning algorithms, so as to leave only |
| 175 | 174 'the equations' to define in particular sub-classes, using Theano. |
| 176 In the default implementations of use and update, it is assumed that the 'use' and 'update' methods | 175 |
| 177 visit examples in the input dataset sequentially. In the 'use' method only one pass through the dataset is done, | 176 In the default implementations of use and update, it is assumed |
| 178 whereas the sub-learner may wish to iterate over the examples multiple times. Subclasses where this | 177 that the 'use' and 'update' methods visit examples in the input |
| 179 basic model is not appropriate can simply redefine update or use. | 178 dataset sequentially. In the 'use' method only one pass through the |
| 180 | 179 dataset is done, whereas the sub-learner may wish to iterate over |
| 180 the examples multiple times. Subclasses where this basic model is | |
| 181 not appropriate can simply redefine update or use. | |
| 182 | |
| 181 Sub-classes must provide the following functions and functionalities: | 183 Sub-classes must provide the following functions and functionalities: |
| 182 - attributeNames(): defines all the names of attributes which can be used as fields or | 184 - attributeNames(): defines all the names of attributes which can |
| 183 attributes in input/output datasets or in stats collectors. | 185 be used as fields or |
| 184 All these attributes are expected to be theano.Result objects | 186 attributes in input/output datasets or in |
| 185 (with a .data property and recognized by theano.Function for compilation). | 187 stats collectors. All these attributes |
| 186 The sub-class constructor defines the relations between | 188 are expected to be theano.Result objects |
| 187 the Theano variables that may be used by 'use' and 'update' | 189 (with a .data property and recognized by |
| 188 or by a stats collector. | 190 theano.Function for compilation). The sub-class |
| 189 - defaultOutputFields(input_fields): return a list of default dataset output fields when | 191 constructor defines the relations between the |
| 192 Theano variables that may be used by 'use' | |
| 193 and 'update' or by a stats collector. | |
| 194 - defaultOutputFields(input_fields): return a list of default | |
| 195 dataset output fields when | |
| 190 None are provided by the caller of use. | 196 None are provided by the caller of use. |
| 191 The following naming convention is assumed and important. | 197 The following naming convention is assumed and important. Attributes |
| 192 Attributes whose names are listed in attributeNames() can be of any type, | 198 whose names are listed in attributeNames() can be of any type, |
| 193 but those that can be referenced as input/output dataset fields or as | 199 but those that can be referenced as input/output dataset fields or |
| 194 output attributes in 'use' or as input attributes in the stats collector | 200 as output attributes in 'use' or as input attributes in the stats |
| 195 should be associated with a Theano Result variable. If the exported attribute | 201 collector should be associated with a Theano Result variable. If the |
| 196 name is <name>, the corresponding Result name (an internal attribute of | 202 exported attribute name is <name>, the corresponding Result name |
| 197 the TLearner, created in the sub-class constructor) should be _<name>. | 203 (an internal attribute of the TLearner, created in the sub-class |
| 198 Typically <name> will be numpy ndarray and _<name> will be the corresponding | 204 constructor) should be _<name>. Typically <name> will be numpy |
| 199 Theano Tensor (for symbolic manipulation). | 205 ndarray and _<name> will be the corresponding Theano Tensor (for |
| 206 symbolic manipulation). | |
| 200 | 207 |
| 201 @todo pousser dans Learner toute la poutine qui peut l'etre sans etre | 208 @todo pousser dans Learner toute la poutine qui peut l'etre sans etre |
| 202 dependant de Theano | 209 dependant de Theano |
| 203 """ | 210 """ |
| 204 | 211 |
| 250 return [self.__getattribute__('_'+name) for name in names] | 257 return [self.__getattribute__('_'+name) for name in names] |
| 251 | 258 |
| 252 | 259 |
| 253 class MinibatchUpdatesTLearner(TLearner): | 260 class MinibatchUpdatesTLearner(TLearner): |
| 254 """ | 261 """ |
| 255 This adds to L{TLearner} a | 262 This adds the following functions to a L{TLearner}: |
| 256 - updateStart(), updateEnd(), updateMinibatch(minibatch), isLastEpoch(): | 263 - updateStart(), updateEnd(), updateMinibatch(minibatch), isLastEpoch(): |
| 257 functions executed at the beginning, the end, in the middle | 264 functions executed at the beginning, the end, in the middle (for |
| 258 (for each minibatch) of the update method, and at the end | 265 each minibatch) of the update method, and at the end of each |
| 259 of each epoch. This model only | 266 epoch. This model only works for 'online' or one-shot learning |
| 260 works for 'online' or one-shot learning that requires | 267 that requires going only once through the training data. For more |
| 261 going only once through the training data. For more complicated | 268 complicated models, more specialized subclasses of TLearner should |
| 262 models, more specialized subclasses of TLearner should be used | 269 be used or a learning-algorithm specific update method should |
| 263 or a learning-algorithm specific update method should be defined. | 270 be defined. |
| 264 | 271 |
| 265 - a 'parameters' attribute which is a list of parameters (whose names are | 272 - a 'parameters' attribute which is a list of parameters |
| 266 specified by the user's subclass with the parameterAttributes() method) | 273 (whose names are specified by the user's subclass with the |
| 267 | 274 parameterAttributes() method) |
| 275 | |
| 268 """ | 276 """ |
| 269 | 277 |
| 270 def __init__(self): | 278 def __init__(self): |
| 271 TLearner.__init__(self) | 279 TLearner.__init__(self) |
| 272 self.update_minibatch_function = compile.function(self.names2OpResults(self.updateMinibatchOutputAttributes()+ | 280 self.update_minibatch_function = compile.function(self.names2OpResults(self.updateMinibatchOutputAttributes()+ |