download=True,
transform=transform)
test_dataset = datasets.MNIST(data_dir,
train=False,
download=False,
transform=transform)
e.setup_dataloader((train_dataset, None, test_dataset))
# Plot a subset of the training dataset
utils.plot_data_subset(e.fname("dataset_image.png"), train_dataset)
# Setup the two models
e.generator = models.densegan4_generator(e)
e.discriminator = models.densegan4_discriminator(e)
# Criterion (or loss function) used
e.criterion = nn.BCELoss()
# The optimizer for weight updating
e.g_optimizer = optim.Adam(e.generator.parameters(),
lr=e.params["lr"],
betas=e.params["betas"])
e.d_optimizer = optim.Adam(e.discriminator.parameters(),
lr=e.params["lr"],
betas=e.params["betas"])
# Train model and plot results
gantraining.train_model(
e, flatten_input=True) # This network requires flattened images.
gantraining.plot_all(e)
e.params["betas"] = (0.5, 0.999) # Betas hyperparameter for Adam optimizers
e.params["patience"] = 7 # Number of epochs to wait before early stopping
# Setup the CIFAR10 dataset
transform = utils.image_transform(e.params["im_size"])
data_dir = "data/celeba/"
# WARNING DOWNLOAD IS 1.4 GB in size!!!
# train_dataset = datasets.CelebA(data_dir, split="train", download=False, transform=transform)
train_dataset = datasets.ImageFolder(data_dir, transform=transform)
e.setup_dataloader((train_dataset, None, None))
# Plot a subset of the training dataset
utils.plot_data_subset(e.fname("dataset_image.png"), train_dataset, show_labels=False)
# Setup the two models
e.generator = models.dcgan4_generator(e)
e.discriminator = models.dcgan4_discriminator(e)
# Criterion (or loss function) used
e.criterion = nn.BCELoss()
# The optimizer for weight updating
e.g_optimizer = optim.Adam(e.generator.parameters(), lr=e.params["lr"], betas=e.params["betas"])
e.d_optimizer = optim.Adam(e.discriminator.parameters(), lr=e.params["lr"], betas=e.params["betas"])
# Train model and plot results
gantraining.train_model(e)
gantraining.plot_all(e)