def analyse(data_loader, plot=False):
bpds = []
batch_idx = 0
with torch.no_grad():
for data, _ in data_loader:
batch_idx += 1
if args.cuda:
data = data.cuda()
data = data.view(-1, *args.input_size)
loss, batch_bpd, bpd_per_prior, pz, z, pys, ys, ldj = \
model(data)
loss = torch.mean(loss).item()
batch_bpd = torch.mean(batch_bpd).item()
bpds.append(batch_bpd)
bpd = np.mean(bpds)
with torch.no_grad():
if not testing and plot:
x_sample = model_sample.sample(n=100)
try:
plot_reconstructions(x_sample, bpd, loss_type, epoch, args)
except:
print('Not plotting')
return bpd
def analyse(data_loader, plot=False):
bpds = []
losses = []
with torch.no_grad():
for batch_idx, (data, ) in enumerate(data_loader):
if args.device == 'cuda':
data = data.cuda()
data = data.view(-1, *args.input_size)
data = data.to(dtype=torch.float64, device=args.device)
result = model(data)
loss = -torch.mean(result['total_logd'])
bpd = -torch.mean(result['total_logd']) / (64 * 64 * 3)
batch_loss = -torch.mean(result['total_logd']).item()
batch_bpd = (-torch.mean(result['total_logd']) /
(64 * 64 * 3)).item()
if not np.isnan(batch_bpd):
bpds.append(batch_bpd)
losses.append(batch_loss)
tmp = 'Validation Epoch: {:3d} \tLoss: {:11.6f}\tbpd: {:8.6f}'
print(tmp.format(epoch, batch_loss, batch_bpd))
bpd = np.mean(bpds)
loss = np.mean(losses)
with torch.no_grad():
if not testing and plot:
x_sample = model_sample.sample(n=100)
try:
plot_reconstructions(x_sample, bpd, loss_type, epoch, args)
except:
print('Not plotting')
return bpd, loss
def evaluate(data_loader, model, args, testing=False, file=None, epoch=0):
model.eval()
loss = 0.
batch_idx = 0
bpd = 0.
if args.input_type == 'binary':
loss_type = 'elbo'
else:
loss_type = 'bpd'
with torch.no_grad():
for data, _ in data_loader:
batch_idx += 1
if args.cuda:
data = data.cuda()
data = Variable(data)
data = data.view(-1, *args.input_size)
x_mean, z_mu, z_var, ldj, z0, zk = model(data)
batch_loss, rec, kl, batch_bpd = calculate_loss(
x_mean, data, z_mu, z_var, z0, zk, ldj, args)
bpd += batch_bpd
# loss += batch_loss.data[0]
loss += batch_loss.data.item()
# PRINT RECONSTRUCTIONS
if batch_idx == 1 and testing is False:
plot_reconstructions(data, x_mean, batch_loss, loss_type,
epoch, args)
loss /= len(data_loader)
bpd /= len(data_loader)
# Compute log-likelihood
if testing:
test_data = Variable(data_loader.dataset.data_tensor, volatile=True)
if args.cuda:
test_data = test_data.cuda()
print('Computing log-likelihood on test set')
model.eval()
if args.dataset == 'caltech':
log_likelihood, nll_bpd = calculate_likelihood(test_data,
model,
args,
S=2000,
MB=1000)
else:
log_likelihood, nll_bpd = calculate_likelihood(test_data,
model,
args,
S=5000,
MB=1000)
else:
log_likelihood = None
nll_bpd = None
if args.input_type in ['multinomial']:
bpd = loss / (np.prod(args.input_size) * np.log(2.))
if file is None:
if testing:
print('====> Test set loss: {:.4f}'.format(loss))
print(
'====> Test set log-likelihood: {:.4f}'.format(log_likelihood))
if args.input_type != 'binary':
print('====> Test set bpd (elbo): {:.4f}'.format(bpd))
print('====> Test set bpd (log-likelihood): {:.4f}'.format(
log_likelihood / (np.prod(args.input_size) * np.log(2.))))
else:
print('====> Validation set loss: {:.4f}'.format(loss))
if args.input_type in ['multinomial']:
print('====> Validation set bpd: {:.4f}'.format(bpd))
else:
with open(file, 'a') as ff:
if testing:
print('====> Test set loss: {:.4f}'.format(loss), file=ff)
print('====> Test set log-likelihood: {:.4f}'.format(
log_likelihood),
file=ff)
if args.input_type != 'binary':
print('====> Test set bpd: {:.4f}'.format(bpd), file=ff)
print('====> Test set bpd (log-likelihood): {:.4f}'.format(
log_likelihood /
(np.prod(args.input_size) * np.log(2.))),
file=ff)
else:
print('====> Validation set loss: {:.4f}'.format(loss),
file=ff)
if args.input_type != 'binary':
print('====> Validation set bpd: {:.4f}'.format(
loss / (np.prod(args.input_size) * np.log(2.))),
file=ff)
if not testing:
return loss, bpd
else:
return log_likelihood, nll_bpd