def test_score_errors_event_dim_mismatch(dist):
for idx in dist.get_batch_data_indices():
dist_params = dist.get_dist_params(idx)
d = dist.pyro_dist(**dist_params)
test_data_wrong_dims = torch.ones(d.shape() + (1,))
if len(d.event_shape) > 0:
if dist.get_test_distribution_name() == 'MultivariateNormal':
pytest.skip('MultivariateNormal does not do shape validation in log_prob.')
if dist.get_test_distribution_name() == 'LowRankMultivariateNormal':
pytest.skip('LowRankMultivariateNormal does not do shape validation in log_prob.')
with pytest.raises((ValueError, RuntimeError)):
d.log_prob(test_data_wrong_dims)
def test_score_errors_event_dim_mismatch(dist):
for idx in dist.get_batch_data_indices():
dist_params = dist.get_dist_params(idx)
d = dist.pyro_dist(**dist_params)
test_data_wrong_dims = torch.ones(d.shape() + (1,))
if len(d.event_shape) > 0:
if dist.get_test_distribution_name() == 'MultivariateNormal':
pytest.skip('MultivariateNormal does not do shape validation in log_prob.')
elif dist.get_test_distribution_name() == 'LowRankMultivariateNormal':
pytest.skip('LowRankMultivariateNormal does not do shape validation in log_prob.')
with pytest.raises((ValueError, RuntimeError)):
d.log_prob(test_data_wrong_dims)
def test_expand_by(dist, sample_shape, shape_type):
for idx in range(dist.get_num_test_data()):
small = dist.pyro_dist(**dist.get_dist_params(idx))
large = small.expand_by(shape_type(sample_shape))
assert large.batch_shape == sample_shape + small.batch_shape
if dist.get_test_distribution_name() == 'Stable':
pytest.skip('Stable does not implement a log_prob method.')
check_sample_shapes(small, large)
def test_score_errors_non_broadcastable_data_shape(dist):
for idx in dist.get_batch_data_indices():
dist_params = dist.get_dist_params(idx)
d = dist.pyro_dist(**dist_params)
if dist.get_test_distribution_name() == "LKJCholesky":
pytest.skip("https://github.com/pytorch/pytorch/issues/52724")
shape = d.shape()
non_broadcastable_shape = (shape[0] + 1, ) + shape[1:]
test_data_non_broadcastable = torch.ones(non_broadcastable_shape)
with pytest.raises((ValueError, RuntimeError)):
d.log_prob(test_data_non_broadcastable)
def test_expand_error(dist, initial_shape, proposed_shape):
for idx in range(dist.get_num_test_data()):
small = dist.pyro_dist(**dist.get_dist_params(idx))
with xfail_if_not_implemented():
large = small.expand(torch.Size(initial_shape) + small.batch_shape)
proposed_batch_shape = torch.Size(proposed_shape) + small.batch_shape
if dist.get_test_distribution_name() == 'LKJCorrCholesky':
pytest.skip('LKJCorrCholesky can expand to a shape not' +
'broadcastable with its original batch_shape.')
with pytest.raises(RuntimeError):
large.expand(proposed_batch_shape)
def test_expand_new_dim(dist, sample_shape, shape_type, default):
for idx in range(dist.get_num_test_data()):
small = dist.pyro_dist(**dist.get_dist_params(idx))
if default:
large = TorchDistribution.expand(small, shape_type(sample_shape + small.batch_shape))
else:
with xfail_if_not_implemented():
large = small.expand(shape_type(sample_shape + small.batch_shape))
assert large.batch_shape == sample_shape + small.batch_shape
if dist.get_test_distribution_name() == 'Stable':
pytest.skip('Stable does not implement a log_prob method.')
check_sample_shapes(small, large)
def test_expand_existing_dim(dist, shape_type, default):
for idx in range(dist.get_num_test_data()):
small = dist.pyro_dist(**dist.get_dist_params(idx))
for dim, size in enumerate(small.batch_shape):
if size != 1:
continue
batch_shape = list(small.batch_shape)
batch_shape[dim] = 5
batch_shape = torch.Size(batch_shape)
if default:
large = TorchDistribution.expand(small, shape_type(batch_shape))
else:
with xfail_if_not_implemented():
large = small.expand(shape_type(batch_shape))
assert large.batch_shape == batch_shape
if dist.get_test_distribution_name() == 'Stable':
pytest.skip('Stable does not implement a log_prob method.')
check_sample_shapes(small, large)
