# HG changeset patch
# User SylvainPL <sylvain.pannetier.lebeuf@umontreal.ca>
# Date 1270143844 14400
# Node ID a5abd5c8b4b05add688905726c7b350e6724e951
# Parent  60cacb9a70e4bb9d91958795f8e79f98dab5d66e
Petits changements pour pouvoir utiliser le GPU

diff -r 60cacb9a70e4 -r a5abd5c8b4b0 deep/stacked_dae/v_sylvain/sgd_optimization.py
--- a/deep/stacked_dae/v_sylvain/sgd_optimization.py	Thu Apr 01 13:43:43 2010 -0400
+++ b/deep/stacked_dae/v_sylvain/sgd_optimization.py	Thu Apr 01 13:44:04 2010 -0400
@@ -19,6 +19,9 @@
 
 from ift6266.utils.seriestables import *
 
+#For test purpose only
+buffersize=1000
+
 default_series = { \
         'reconstruction_error' : DummySeries(),
         'training_error' : DummySeries(),
@@ -203,7 +206,7 @@
         while (epoch < num_finetune) and (not done_looping):
             epoch = epoch + 1
 
-            for x,y in dataset.train(minibatch_size):
+            for x,y in dataset.train(minibatch_size,bufsize=buffersize):
                 minibatch_index += 1
                 if special == 0:
                     cost_ij = self.classifier.finetune(x,y)
@@ -217,9 +220,9 @@
                     #minibatch_index += 1
                     #The validation set is always NIST (we want the model to be good on NIST)
                     if ind_test == 0 | ind_test == 20:
-                        iter=dataset_test.valid(minibatch_size)
+                        iter=dataset_test.valid(minibatch_size,bufsize=buffersize)
                     else:
-                        iter = dataset.valid(minibatch_size)
+                        iter = dataset.valid(minibatch_size,bufsize=buffersize)
                     if self.max_minibatches:
                         iter = itermax(iter, self.max_minibatches)
                     validation_losses = [validate_model(x,y) for x,y in iter]
@@ -247,14 +250,14 @@
                         parameters_finetune=[copy(x.value) for x in self.classifier.params]
 
                         # test it on the test set
-                        iter = dataset.test(minibatch_size)
+                        iter = dataset.test(minibatch_size,bufsize=buffersize)
                         if self.max_minibatches:
                             iter = itermax(iter, self.max_minibatches)
                         test_losses = [test_model(x,y) for x,y in iter]
                         test_score = numpy.mean(test_losses)
                         
                         #test it on the second test set
-                        iter2 = dataset_test.test(minibatch_size)
+                        iter2 = dataset_test.test(minibatch_size,bufsize=buffersize)
                         if self.max_minibatches:
                             iter2 = itermax(iter2, self.max_minibatches)
                         test_losses2 = [test_model(x,y) for x,y in iter2]
@@ -334,7 +337,10 @@
         self.parameters_pre=pickle.load(f)
         f.close()
         for idx,x in enumerate(self.parameters_pre):
-            self.classifier.params[idx].value=copy(x)
+            if x.dtype=='float64':
+                self.classifier.params[idx].value=theano._asarray(copy(x),dtype=theano.config.floatX)
+            else:
+                self.classifier.params[idx].value=copy(x)