params search

actions/train.py

@@ -11,19 +11,21 @@ from tqdm import tqdm
 class Train_Gans:
     def __init__(self, args):
         self.args = args
-        optimizer = Adam(args.lr, args.decay)
+        D_optimizer = Adam(args.d_lr, args.d_decay)
+        G_optimizer = Adam(args.g_lr, args.g_decay)
 
         # Baseline model
         self.baseline_model = build_generator(args, beta=1)
         self.baseline_model.compile(loss='categorical_crossentropy',
-                                    optimizer=optimizer,
+                                    optimizer=D_optimizer,
                                     metrics=['accuracy'])
 
         # Build and compile the discriminator
         self.discriminator = build_discriminator(args)
         self.discriminator.compile(loss='binary_crossentropy',
-                                   optimizer=optimizer,
+                                   optimizer=D_optimizer,
                                    metrics=['accuracy'])
+
         # Build the generator
         self.generator = build_generator(args)
         input_data = Input(shape=args.img_shape) if args.conv else Input(shape=(np.prod(args.img_shape),))
@@ -32,7 +34,7 @@ class Train_Gans:
         valid = self.discriminator([input_data, fake_labels])
         self.combined = Model(input_data, valid)
         self.combined.compile(loss=['binary_crossentropy'],
-                              optimizer=optimizer, metrics=['accuracy'])
+                              optimizer=G_optimizer, metrics=['accuracy'])
         self.history = {'Loss': {'Discriminator': [], 'Generator': []},
                         'Accuracy': {'Discriminator': [], 'Generator': []}
                         }
@@ -57,7 +59,7 @@ class Train_Gans:
                 d_loss_fake = self.discriminator.train_on_batch([valid_imgs, gen_labels], fake)
                 d_loss = 0.5 * np.add(d_loss_real, d_loss_fake)
             g_loss = self.combined.train_on_batch(valid_imgs, valid)
-            desc = "Epoch %d/%d [D loss: %f, acc.: %.2f%%] [G loss: %f, acc.: %.2f%%]" \
+            desc = "Epoch %d/%d [D loss: %f, acc.: %.2f%%] [G loss: %f, acc.: %.2f%%] " \
                    % (epoch, self.args.epochs, d_loss[0], 100 * d_loss[1], g_loss[0], 100 * g_loss[1])
             self.history['Loss']['Discriminator'].append(d_loss[0])
             self.history['Accuracy']['Discriminator'].append(d_loss[1])