Mercurial > pylearn

comparison mlp_factory_approach.py @ 304:6ead65d30f1e

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
while learning using __call__, we can now set the early stopper
author Thierry Bertin-Mahieux <bertinmt@iro.umontreal.ca>
date Tue, 10 Jun 2008 17:16:49 -0400
parents eded3cb54930
children 93280a0c151a
comparison
equal deleted inserted replaced
303:410a6ef674ed 304:6ead65d30f1e
3 3
4 import theano 4 import theano
5 from theano import tensor as T 5 from theano import tensor as T
6 6
7 import dataset, nnet_ops, stopper, filetensor 7 import dataset, nnet_ops, stopper, filetensor
8 from lookup_list import LookupList 8 from pylearn.lookup_list import LookupList
9 9
10 10
11 class AbstractFunction (Exception): pass 11 class AbstractFunction (Exception): pass
12 12
13 class AutoName(object): 13 class AutoName(object):
121 121
122 def linker(self): 122 def linker(self):
123 return theano.gof.PerformLinker() 123 return theano.gof.PerformLinker()
124 124
125 def early_stopper(self): 125 def early_stopper(self):
126 stopper.NStages(10,1) 126 stopper.NStages(300,1)
127 127
128 def train_iter(self, trainset): 128 def train_iter(self, trainset):
129 raise AbstractFunction 129 raise AbstractFunction
130 optimizer = Opt() 130 optimizer = Opt()
131 131
144 # prefer caching in Model.__call__ 144 # prefer caching in Model.__call__
145 return theano.function(inputs, outputs, 145 return theano.function(inputs, outputs,
146 unpack_single=False, 146 unpack_single=False,
147 optimizer=self.graph.optimizer, 147 optimizer=self.graph.optimizer,
148 linker=self.graph.linker() if hasattr(self.graph, 'linker') 148 linker=self.graph.linker() if hasattr(self.graph, 'linker')
149 else 'c&py') 149 else 'c|py')
150 150
151 def __call__(self, 151 def __call__(self,
152 trainset=None, 152 trainset=None,
153 validset=None, 153 validset=None,
154 iparams=None): 154 iparams=None,
155 stp=None):
155 """Allocate and optionally train a model 156 """Allocate and optionally train a model
156 157
157 @param trainset: Data for minimizing the cost function 158 @param trainset: Data for minimizing the cost function
158 @type trainset: None or Dataset 159 @type trainset: None or Dataset
159 160
163 @param input: name of field to use as input 164 @param input: name of field to use as input
164 @type input: string 165 @type input: string
165 166
166 @param target: name of field to use as target 167 @param target: name of field to use as target
167 @type target: string 168 @type target: string
169
170 @param stp: early stopper, if None use default in graphMLP.G
171 @type stp: None or early stopper
168 172
169 @return: model 173 @return: model
170 @rtype: GraphLearner.Model instance 174 @rtype: GraphLearner.Model instance
171 175
172 """ 176 """
182 curmodel = GraphLearner.Model(self, iparams) 186 curmodel = GraphLearner.Model(self, iparams)
183 best = curmodel 187 best = curmodel
184 188
185 if trainset is not None: 189 if trainset is not None:
186 #do some training by calling Model.update_minibatch() 190 #do some training by calling Model.update_minibatch()
187 stp = self.graph.early_stopper() 191 if stp == None :
188 for mb in self.graph.train_iter(trainset): 192 stp = self.graph.early_stopper()
189 curmodel.update_minibatch(mb) 193 try :
190 if stp.set_score: 194 countiter = 0
191 if validset: 195 for mb in self.graph.train_iter(trainset):
192 stp.score = curmodel(validset, ['validset_score']) 196 curmodel.update_minibatch(mb)
193 if (stp.score < stp.best_score): 197 if stp.set_score:
194 best = copy.copy(curmodel) 198 if validset:
195 else: 199 stp.score = curmodel(validset, ['validset_score'])
196 stp.score = 0.0 200 if (stp.score < stp.best_score):
197 stp.next() 201 best = copy.copy(curmodel)
202 else:
203 stp.score = 0.0
204 countiter +=1
205 stp.next()
206 except StopIteration :
207 print 'Iterations stopped after ', countiter,' iterations'
198 if validset: 208 if validset:
199 curmodel = best 209 curmodel = best
200 return curmodel 210 return curmodel
201 211
202 212
276 , randsmall(nhid, nclass) 286 , randsmall(nhid, nclass)
277 , randsmall(nclass)] 287 , randsmall(nclass)]
278 288
279 def train_iter(self, trainset): 289 def train_iter(self, trainset):
280 return trainset.minibatches(['input', 'target'], 290 return trainset.minibatches(['input', 'target'],
281 minibatch_size=min(len(trainset), 32), n_batches=300) 291 minibatch_size=min(len(trainset), 32), n_batches=2000)
282 def early_stopper(self): 292 def early_stopper(self):
293 """ overwrites GraphLearner.graph function """
283 return stopper.NStages(300,1) 294 return stopper.NStages(300,1)
284 295
285 return G() 296 return G()
286 297
287 298