def main():
args = get_args()
if not args.silent:
save_path = os.path.abspath(script_path + args.save_path)
if not os.path.exists(save_path):os.mkdir(save_path)
save_path = os.path.abspath(save_path + "/" + args.name)
if not os.path.exists(save_path):os.mkdir(save_path)
preview_path = os.path.abspath(save_path + "/preview")
if not os.path.exists(preview_path):os.mkdir(preview_path)
dataset = Dataset(args)
if args.max_epoch is not None:
epoch_iter = dataset.train_data_len // args.batch_size
if dataset.train_data_len % args.batch_size != 0:epoch_iter += 1
args.max_iter = args.max_epoch * epoch_iter
progress = print_progress(args.max_iter, args.batch_size, dataset.train_data_len)
if args.gpu_num != 0:
cuda.get_device_from_array(xp.array([i for i in range(args.gpu_num)])).use()
model = make_model(args, dataset)
netG_opt = make_optimizer(model.netG_0, args.adam_alpha, args.adam_beta1, args.adam_beta2)
netD_opt = make_optimizer(model.netD_0, args.adam_alpha, args.adam_beta1, args.adam_beta2)
updater = Updater(model, netG_opt, netD_opt, args.n_dis, args.batch_size, args.gpu_num, args.KL_loss_iter, args.KL_loss_conf, args.epoch_decay, args.max_iter)
print("==========================================")
print("Info:start train")
start = time.time()
for i in range(args.max_iter):
data = toGPU(dataset.next(), args.gpu_num)
updater.update(data, dataset.now_epoch)
if dataset.now_iter % args.display_interval == 0:
elapsed = time.time() - start
progress(elapsed, dataset.get_state)
np.save(save_path + "/loss_hist.npy", updater.loss_hist)
start = time.time()
if dataset.now_iter % args.snapshot_interval == 0 and not args.silent:
data = dataset.sampling(args.sample_size)
sample = sample_generate(model.netG_0, data, args.noise_dim, args.noise_dist)
Image.fromarray(sample).save(preview_path + f"/image_{dataset.now_iter:08d}.png")
serializers.save_npz(save_path + f"/Generator_{dataset.now_iter:08d}.npz",model.netG_0)
serializers.save_npz(save_path + f"/Discriminator_{dataset.now_iter:08d}.npz",model.netD_0)
if not args.silent:
data = dataset.sampling(args.sample_size)
sample = sample_generate(model.netG_0, data, args.noise_dim, args.noise_dist)
Image.fromarray(sample).save(preview_path + f"/image_{dataset.now_iter:08d}.png")
serializers.save_npz(save_path + f"/Generator_{dataset.now_iter:08d}.npz",model.netG_0)
serializers.save_npz(save_path + f"/Discriminator_{dataset.now_iter:08d}.npz",model.netD_0)
print("\n\n\n\n==========================================")
print("Info:finish train")
if batch_idx == 0:
visualize_results(epoch, inputs, outputs)
# Save checkpoint.
test_loss = test_loss / (batch_idx + 1)
writer.add_scalar('test/loss', test_loss, epoch)
return test_loss
#
# main script
#
# get command-line arguments
args = get_args()
args.network = 'deep_autoencoder'
args.dataset = 'mnist'
args.activation = 'tanh'
# set random seed for reproducibility
torch.manual_seed(args.seed)
# choose activation fuction for model
act_dict = {
'relu': torch.nn.ReLU(),
'sigmoid': torch.nn.Sigmoid(),
'tanh': torch.nn.Tanh()
}
act = torch.nn.Sigmoid() if args.activation is None else act_dict[
def select_gcard(args):
filename = fs.dirname + "/valid_gcards.txt"
with open(filename) as f:
content = f.readlines()
content = [x.strip("\n") for x in content]
for linenum, line in enumerate(content):
if not linenum == 0:
print("({0}) - {1}".format(linenum, line))
else:
print(line + "\n")
selection = selector(content)
while (selection not in np.arange(1, len(content))
or not (isinstance(selection, int))):
print(("\n Selection not in valid range, try again, or "
"hit ctrl+c to exit"))
selection = selector(content)
gcard_selected = content[selection].split(',')[0]
print("Gcard for simultions will be {0}".format(gcard_selected))
return gcard_selected
if __name__ == "__main__":
args = get_args.get_args()
select_gcard(args)