def test_convolution(dist, n_dims):
assert dist.sample(1).shape == (1, n_dims)
assert dist.sample(64).shape == (64, n_dims)
try:
dist.log_prob(dist.sample(64))
except NotImplementedError:
pass
dist.get_parameters()
def test_gumbel_softmax(dist, n_dims):
assert dist.sample(1).shape == (1, n_dims)
assert dist.log_prob(dist.sample(1)).shape == (1, )
samples = dist.sample(64)
assert samples.shape == (64, n_dims)
log_probs = dist.log_prob(samples)
assert log_probs.shape == (64, )
dist.get_parameters()
try:
dist.entropy()
except NotImplementedError:
pass
def test_normal_dist(dist, n_dims):
assert dist.sample(1).shape == (1, n_dims)
assert dist.log_prob(dist.sample(1)).shape == (1, )
samples = dist.sample(64)
assert samples.shape == (64, n_dims)
log_probs = dist.log_prob(samples)
assert log_probs.shape == (64, )
dist.get_parameters()
dist.num_parameters
try:
dist.entropy()
except NotImplementedError:
pass
try:
dist.perplexity()
except NotImplementedError:
pass
def test_dirac_delta(loc, n_dims):
dist = DiracDelta(loc)
assert dist.sample(1).shape == (1, n_dims), "{} {}".format(
dist.sample(1).shape, n_dims)
try:
dist.log_prob(dist.sample(64))
except NotImplementedError:
pass
samples = dist.sample(64)
assert samples.shape == (64, n_dims)
if n_dims == 1:
assert dist.get_parameters()['loc'] == 1.0
elif n_dims == 2:
assert (dist.get_parameters()['loc'] == np.array([2.0, 3.0])).all()
elif n_dims == 3:
assert (dist.get_parameters()['loc'] == np.array([1.0, 2.0,
3.0])).all()
else:
raise ValueError()
dist.get_parameters()
def test_data_dist(n_dims):
data = torch.randn(1000, n_dims)
dist = Data(data)
assert dist.sample(1).shape == (1, n_dims)
assert dist.sample(64).shape == (64, n_dims)
try:
dist.log_prob(dist.sample(64))
except NotImplementedError:
pass
assert dist.get_parameters()['n_dims'] == n_dims
assert dist.get_parameters()['n_samples'] == 1000
data = np.random.randn(100, n_dims)
dist = Data(data)
assert dist.sample(1).shape == (1, n_dims)
assert dist.sample(64).shape == (64, n_dims)
assert dist.get_parameters()['n_dims'] == n_dims
assert dist.get_parameters()['n_samples'] == 100
def sampleS(dist):
print('sample* is called')
print('input to sample is: ', str(dist))
# takes in distribution type variable
return dist.sample()
def test_dirac_delta_2d():
dist = DiracDelta([[1, 0], [0, 1]])
assert dist.sample(1).shape == (1, 2, 2), "{} {}".format(
dist.sample(1).shape, n_dims)
dist.get_parameters()
def reparameterize(self, dist, reparam=True):
if reparam:
latent = dist.rsample()
else:
latent = dist.sample()
return latent