def generate(self, num_to_sample: int = 1):
cuda_device = self._get_prediction_device()
prior_mean = nn_util.move_to_device(
torch.zeros((num_to_sample, self._latent_dim)), cuda_device)
prior_stddev = torch.ones_like(prior_mean)
prior = Normal(prior_mean, prior_stddev)
latent = prior.sample()
generated = self._decoder.generate(latent)
return self.decode(generated)
def test_mask_type(mask):
p = Normal(torch.randn(2, 2), torch.randn(2, 2).exp())
p_masked = p.mask(mask)
if isinstance(mask, bool):
mask = torch.tensor(mask)
x = p.sample()
actual = p_masked.log_prob(x)
expected = p.log_prob(x) * mask.float()
assert_equal(actual, expected)
actual = p_masked.score_parts(x)
expected = p.score_parts(x)
for a, e in zip(actual, expected):
if isinstance(e, torch.Tensor):
e = e * mask.float()
assert_equal(a, e)