nc=args.nc,
model_type=args.model_type)
src_loaders = load_datasets(args=args)
loss = wae.train_model(model, src_loaders['train'], src_loaders['val'],
device, args)
# conditional generation
model.eval()
evaluation.sampling(model, device, args.epochs, args, prefix='wae', nrow=4)
# t-sne visualization
if args.source_data == 'MNIST':
evaluation.visualization_tsne(model,
src_loaders['val'],
device,
args,
prefix='wae')
else:
evaluation.visualization_tsne2(model,
src_loaders['val'],
device,
args,
prefix='wae')
# save models and learning results
model = model.to('cpu')
torch.save(
model.state_dict(),
'{}/wae_model_{}_{}.pt'.format(args.resultpath, args.model_type,
args.source_data))
prior_list = [z_p_mean, z_p_logvar]
for i in range(args.K):
evaluation.sampling(
model,
device,
i + 1,
args,
prefix='rae',
prior=[z_p_mean[i, :].unsqueeze(0), z_p_logvar[i, :].unsqueeze(0)],
nrow=4)
# t-sne visualization
if args.source_data == 'MNIST':
evaluation.visualization_tsne(model,
src_loaders['val'],
device,
args,
prefix='rae',
prior=prior_list)
else:
evaluation.visualization_tsne2(model,
src_loaders['val'],
device,
args,
prefix='rae',
prior=prior_list)
# save models and learning results
model = model.to('cpu')
prior = prior.to('cpu')
torch.save(
model.state_dict(),