Gan conv for 32X32.

ragavvenkatesan · Mar 11, 2017 · 9806a9a · 9806a9a
1 parent e2e6353
commit 9806a9a
Show file tree

Hide file tree

Showing 3 changed files with 23 additions and 20 deletions.
diff --git a/pantry/matlab/make_svhn.m b/pantry/matlab/make_svhn.m
@@ -57,10 +57,10 @@
 labels = y (1:length(y) - throw_away) - 1; % because labels go from 1-10
 
 total_batches = length(labels) / batch_size;
-test_size = total_batches / 3;
+test_size = 130;
 remain = total_batches - test_size; 
 
-train_size = 2* remain / 3;
+train_size = 1000;
 remain = remain - train_size;
 valid_size = remain; 
 

diff --git a/pantry/tutorials/gan.py b/pantry/tutorials/gan.py
@@ -1,5 +1,5 @@
 """
-Implementation from 
+Referenced from 
 
 Goodfellow, Ian, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, 
 Aaron Courville, and Yoshua Bengio. "Generative adversarial nets." In Advances in Neural Information
@@ -469,6 +469,9 @@ def deep_deconvolutional_gan (dataset,
     This is an example code. You should study this code rather than merely run it.  
     This method uses a few deconvolutional layers as was used in the DCGAN paper.
 
+    This method is setup to produce images of size 32X32. Use a 32X32 dataset such as the Cifar or 
+    the SVHN dataset or change its setup to run for a 28X28 dataset and run MNIST.
+
     Args:         
         dataset: Supply a dataset.    
         regularize: ``True`` (default) supplied to layer arguments
@@ -487,8 +490,8 @@ def deep_deconvolutional_gan (dataset,
 
     optimizer_params =  {        
                 "momentum_type"       : 'nesterov',             
-                "momentum_params"     : (0.9, 0.99, 20),      
-                "regularization"      : (0.0001, 0.0001),       
+                "momentum_params"     : (0.65, 0.95, 10),      
+                "regularization"      : (0.00001, 0.00001),       
                 "optimizer_type"      : 'rmsprop',                
                 "id"                  : "main"
                         }
@@ -504,7 +507,7 @@ def deep_deconvolutional_gan (dataset,
                     "root"       : '.',
                     "frequency"  : 1,
                     "sample_size": 225,
-                    "rgb_filters": False,
+                    "rgb_filters": True,
                     "debug_functions" : False,
                     "debug_layers": False,  
                     "id"         : 'main'
@@ -526,7 +529,7 @@ def deep_deconvolutional_gan (dataset,
     #z - latent space created by random layer
     net.add_layer(type = 'random',
                         id = 'z',
-                        num_neurons = (100,100), 
+                        num_neurons = (100,32), 
                         distribution = 'normal',
                         mu = 0,
                         sigma = 1,
@@ -547,7 +550,7 @@ def deep_deconvolutional_gan (dataset,
     net.add_layer ( type = "dot_product",
                     origin = "G1",
                     id = "G2",
-                    num_neurons = 1440,
+                    num_neurons = 5408,
                     activation = 'relu',
                     regularize = regularize,
                     batch_norm = batch_norm,
@@ -557,17 +560,17 @@ def deep_deconvolutional_gan (dataset,
     net.add_layer ( type = "unflatten",
                     origin = "G2",
                     id = "G2-unflatten",
-                    shape = (12, 12, 10),
+                    shape = (13, 13, 32),
                     batch_norm = batch_norm,
                     verbose = verbose
                     )
 
     net.add_layer ( type = "deconv",
                     origin = "G2-unflatten",
                     id = "G3",
-                    num_neurons = 10,
+                    num_neurons = 32,
                     filter_size = (3,3),
-                    output_shape = (26,26,32),
+                    output_shape = (28,28,32),
                     activation = 'relu',
                     regularize = regularize,    
                     batch_norm = batch_norm,
@@ -579,8 +582,8 @@ def deep_deconvolutional_gan (dataset,
                     origin = "G3",
                     id = "G(z)",
                     num_neurons = 32,
-                    filter_size = (3,3),
-                    output_shape = (28,28,1),
+                    filter_size = (5,5),
+                    output_shape = (32,32,3),
                     activation = 'tanh',
                     # regularize = regularize,    
                     stride = (1,1),
@@ -784,11 +787,11 @@ def deep_deconvolutional_gan (dataset,
                 game_layers = ("D(x)", "D(G(z))"),
                 verbose = verbose )
 
-    learning_rates = (0.00004, 0.01 )  
+    learning_rates = (0.0004, 0.001 )  
 
     net.train( epochs = (20), 
             k = 1, 
-            pre_train_discriminator = 3,
+            pre_train_discriminator = 0,
             validate_after_epochs = 1,
             visualize_after_epochs = 1,
             training_accuracy = True,
@@ -825,4 +828,4 @@ def deep_deconvolutional_gan (dataset,
                                      dropout_rate = 0.5,
                                      regularize = True,
                                      dataset = dataset,
-                                     verbose = 2 )        
+                                     verbose = 2 )        
diff --git a/pantry/tutorials/mat2yann.py b/pantry/tutorials/mat2yann.py
@@ -32,9 +32,9 @@ def cook_svhn_normalized( location, verbose = 1, **kwargs):
                    "height"             : 32,
                    "width"              : 32,
                    "channels"           : 3,
-                   "batches2test"       : 42,
-                   "batches2train"      : 56,
-                   "batches2validate"   : 28,
+                   "batches2test"       : 13,
+                   "batches2train"      : 100,
+                   "batches2validate"   : 13,
                    "mini_batch_size"    : 500}
 
     else:
@@ -47,7 +47,7 @@ def cook_svhn_normalized( location, verbose = 1, **kwargs):
                             "normalize"     : True,
                             "ZCA"           : False,
                             "grayscale"     : False,
-                            "zero_mean"     : False,
+                            "zero_mean"     : True,
                         }
     else:
         preprocess_params = kwargs['preprocess_params']