def make_group_generator():
# Note that this Variable is NOT going to show up in `net.parameters()` and
# therefore it is implicitly free from the ridge penalty/p(theta) prior.
log_sigma = Variable(torch.log(1e-2 * torch.ones(image_size)),
requires_grad=True)
return NormalNet(mu_net=torch.nn.Linear(group_input_dim, image_size),
sigma_net=Lambda(lambda x, log_sigma: torch.exp(
log_sigma.expand(x.size(0), -1)) + 1e-3,
extra_args=(log_sigma, )))
A = torch.Tensor(A)
data.append(A.view(-1))
temp = torch.stack([data[i] for i in range(len(data))])
X = torch.stack([temp for _ in range(num_samples)])
X += 0.05 * torch.randn(X.size())
X = X.transpose(0, 1)
stddev_multiple = 0.1
inference_net = NormalNet(mu_net=nn.Sequential(nn.Linear(dim_h + dim_h,
dim_z)),
sigma_net=torch.nn.Sequential(
nn.Linear(dim_h + dim_h, dim_z),
Lambda(torch.exp),
Lambda(lambda x: x * stddev_multiple + 1e-3)))
def make_group_generator():
# Note that this Variable is NOT going to show up in `net.parameters()` and
# therefore it is implicitly free from the ridge penalty/p(theta) prior.
log_sigma = Variable(torch.log(1e-2 * torch.ones(image_size)),
requires_grad=True)
return NormalNet(mu_net=torch.nn.Sequential(
torch.nn.Tanh(), torch.nn.Linear(group_input_dim, image_size)),
sigma_net=Lambda(lambda x, log_sigma: torch.exp(
log_sigma.expand(x.size(0), -1)) + 1e-3,
extra_args=(log_sigma, )))