wassersteing gan experiment

labml_nn/gan/dcgan/__init__.py

@@ -17,7 +17,7 @@ import torch.nn as nn
 from labml import experiment
 from labml.configs import calculate
 from labml_helpers.module import Module
-from labml_nn.gan.simple_mnist_experiment import Configs
+from labml_nn.gan.original.experiment import Configs
 
 
 class Generator(Module):
@@ -107,7 +107,7 @@ calculate(Configs.discriminator, 'cnn', lambda c: Discriminator().to(c.device))
 
 def main():
     conf = Configs()
-    experiment.create(name='mnist_dcgan', comment='test')
+    experiment.create(name='mnist_dcgan')
     experiment.configs(conf,
                        {'discriminator': 'cnn',
                         'generator': 'cnn',

labml_nn/gan/original/experiment.py

@@ -85,6 +85,7 @@ class Discriminator(Module):
 
 class Configs(MNISTConfigs, TrainValidConfigs):
     device: torch.device = DeviceConfigs()
+    dataset_transforms = 'mnist_gan_transforms'
     epochs: int = 10
 
     is_save_models = True
@@ -146,7 +147,7 @@ class Configs(MNISTConfigs, TrainValidConfigs):
             loss = self.generator_loss(logits)
 
             # Log stuff
-            tracker.add('generated', generated_images[0:5])
+            tracker.add('generated', generated_images[0:6])
             tracker.add("loss.generator.", loss)
 
             # Train
@@ -161,7 +162,7 @@ class Configs(MNISTConfigs, TrainValidConfigs):
 
 
 @option(Configs.dataset_transforms)
-def mnist_transforms():
+def mnist_gan_transforms():
     return transforms.Compose([
         transforms.ToTensor(),
         transforms.Normalize((0.5,), (0.5,))

labml_nn/gan/wasserstein/experiment.py

@@ -1,5 +1,5 @@
-# We import the [simple gan experiment]((simple_mnist_experiment.html) and change the
-# generator and discriminator networks
+# We import the [DCGAN experiment]((../dcgan.html) and change the
+# loss functions
 from labml import experiment
 
 from labml.configs import calculate