implemented LSGAN

pantry/tutorials/gan.py

@@ -463,7 +463,7 @@ def deep_gan_mnist (dataset, verbose = 1 ):
             early_terminate = True,
             verbose = verbose)
 
-def deep_deconvolutional_gan_cifar(dataset,
+def deep_deconvolutional_gan(dataset,
                               regularize = True,
                               batch_norm = True,
                               dropout_rate = 0.5,
@@ -803,7 +803,7 @@ def deep_deconvolutional_gan_cifar(dataset,
 
     return net
 
-def deep_deconvolutional_gan_svhn(dataset,
+def deep_deconvolutional_lsgan(dataset,
                               regularize = True,
                               batch_norm = True,
                               dropout_rate = 0.5,
@@ -825,14 +825,16 @@ def deep_deconvolutional_gan_svhn(dataset,
         net: A Network object.
 
     Notes:
-        This method is setupfor SVHN.
+        This method is setupfor SVHN / CIFAR10.
+        This is an implementation of th least squares GAN with a = 0, b = 1 and c= 1 (equation 9)
+        [1] Least Squares Generative Adversarial Networks, Xudong Mao, Qing Li, Haoran Xie, Raymond Y.K. Lau, Zhen Wang
     """
     if verbose >=2:
         print (".. Creating a GAN network")
 
     optimizer_params =  {        
                 "momentum_type"       : 'polyak',             
-                "momentum_params"     : (0.51, 0.7, 40),      
+                "momentum_params"     : (0.55, 0.9, 20),      
                 "regularization"      : (0.00001, 0.00001),       
                 "optimizer_type"      : 'adagrad',                
                 "id"                  : "main"
@@ -871,7 +873,7 @@ def deep_deconvolutional_gan_svhn(dataset,
     #z - latent space created by random layer
     net.add_layer(type = 'random',
                         id = 'z',
-                        num_neurons = (500,64), 
+                        num_neurons = (500,32), 
                         distribution = 'normal',
                         mu = 0,
                         sigma = 1,
@@ -1077,8 +1079,8 @@ def deep_deconvolutional_gan_svhn(dataset,
     # objective layers 
     # discriminator objective 
     net.add_layer (type = "tensor",
-                    input =  - 0.5 * T.mean(T.log(net.layers['D(x)'].output)) - \
-                                0.5 * T.mean(T.log(1-net.layers['D(G(z))'].output)),
+                    input =  0.5 * T.mean(T.sqr(net.layers['D(x)'].output-1)) + \
+                                0.5 * T.mean(T.sqr(net.layers['D(G(z))'].output)),
                     input_shape = (1,),
                     id = "discriminator_task"
                     )
@@ -1093,7 +1095,7 @@ def deep_deconvolutional_gan_svhn(dataset,
                     )
     #generator objective 
     net.add_layer (type = "tensor",
-                    input =  - 0.5 * T.mean(T.log(net.layers['D(G(z))'].output)),
+                    input =  0.5 * T.mean(T.sqr(net.layers['D(G(z))'].output-1)),
                     input_shape = (1,),
                     id = "objective_task"
                     )
@@ -1129,11 +1131,11 @@ def deep_deconvolutional_gan_svhn(dataset,
                 game_layers = ("D(x)", "D(G(z))"),
                 verbose = verbose )
 
-    learning_rates = (0.04, 0.01 )  
+    learning_rates = (0.04, 0.0001 )  
 
     net.train( epochs = (20), 
             k = 2, 
-            pre_train_discriminator = 2,
+            pre_train_discriminator = 1,
             validate_after_epochs = 1,
             visualize_after_epochs = 1,
             training_accuracy = True,
@@ -1146,8 +1148,8 @@ def deep_deconvolutional_gan_svhn(dataset,
 
 if __name__ == '__main__':
 
-    from yann.special.datasets import cook_mnist_normalized_zero_mean as c 
-    #from yann.special.datasets import cook_cifar10_normalized_zero_mean as c
+    #from yann.special.datasets import cook_mnist_normalized_zero_mean as c 
+    from yann.special.datasets import cook_cifar10_normalized_zero_mean as c
     import sys
 
     dataset = None  
@@ -1165,14 +1167,14 @@ def deep_deconvolutional_gan_svhn(dataset,
         data = c (verbose = 2)
         dataset = data.dataset_location() 
 
-    net = shallow_gan_mnist ( dataset, verbose = 2 )
-    net = deep_gan_mnist ( dataset, verbose = 2 )           
-    net = deep_deconvolutional_gan_cifar ( batch_norm = True,
+    # net = shallow_gan_mnist ( dataset, verbose = 2 )
+    # net = deep_gan_mnist ( dataset, verbose = 2 )           
+    """net = deep_deconvolutional_gan ( batch_norm = True,
                                      dropout_rate = 0.5,
                                      regularize = True,
                                      dataset = dataset,
-                                     verbose = 2 )
-    net = deep_deconvolutional_gan_svhn ( batch_norm = True,
+                                     verbose = 2 )"""
+    net = deep_deconvolutional_lsgan ( batch_norm = True,
                                      dropout_rate = 0.5,
                                      regularize = True,
                                      dataset = dataset,