--- a/deep/stacked_dae/sgd_optimization.py	Tue Mar 02 14:47:18 2010 -0500
+++ b/deep/stacked_dae/sgd_optimization.py	Wed Mar 03 12:51:40 2010 -0500
@@ -60,20 +60,27 @@
         # compute number of minibatches for training, validation and testing
         self.n_train_batches = self.train_set_x.value.shape[0] / self.hp.minibatch_size
         self.n_valid_batches = self.valid_set_x.value.shape[0] / self.hp.minibatch_size
-        self.n_test_batches  = self.test_set_x.value.shape[0]  / self.hp.minibatch_size
+        # remove last batch in case it's incomplete
+        self.n_test_batches  = (self.test_set_x.value.shape[0]  / self.hp.minibatch_size) - 1
 
     def init_classifier(self):
         print "Constructing classifier"
 
+        # we don't want to save arrays in DD objects, so
+        # we recreate those arrays here
+        nhl = self.hp.num_hidden_layers
+        layers_sizes = [self.hp.hidden_layers_sizes] * nhl
+        corruption_levels = [self.hp.corruption_levels] * nhl
+
         # construct the stacked denoising autoencoder class
         self.classifier = SdA( \
                           train_set_x= self.train_set_x, \
                           train_set_y = self.train_set_y,\
                           batch_size = self.hp.minibatch_size, \
                           n_ins= self.n_ins, \
-                          hidden_layers_sizes = self.hp.hidden_layers_sizes, \
+                          hidden_layers_sizes = layers_sizes, \
                           n_outs = self.n_outs, \
-                          corruption_levels = self.hp.corruption_levels,\
+                          corruption_levels = corruption_levels,\
                           rng = self.rng,\
                           pretrain_lr = self.hp.pretraining_lr, \
                           finetune_lr = self.hp.finetuning_lr,\