def test_conditional_loglikelihood(self):
mcgsm = MCGSM(3, 1, 2, 1, 4)
mcgsm.linear_features = randn(mcgsm.num_components, mcgsm.dim_in) / 5.
mcgsm.means = randn(mcgsm.dim_out, mcgsm.num_components) / 5.
M = 100
inputs = randn(mcgsm.dim_in, M)
outputs = mcgsm.sample(inputs)
loglik0 = mcgsm.loglikelihood(inputs, outputs)
loglik1 = []
N = 1000
# estimate log-likelihood via sampling
for _ in range(N):
labels = mcgsm.sample_prior(inputs)
loglik1.append(mcgsm.loglikelihood(inputs, outputs, labels))
loglik1 = vstack(loglik1)
d = abs(logmeanexp(loglik1, 0) - loglik0).ravel()
s = std(loglik1, 0, ddof=1).ravel()
for i in range(M):
self.assertLess(d[i], 6. * s[i] / sqrt(N))
def test_conditional_loglikelihood(self):
mcgsm = MCGSM(3, 1, 2, 1, 4)
mcgsm.linear_features = randn(mcgsm.num_components, mcgsm.dim_in) / 5.0
mcgsm.means = randn(mcgsm.dim_out, mcgsm.num_components) / 5.0
M = 100
inputs = randn(mcgsm.dim_in, M)
outputs = mcgsm.sample(inputs)
loglik0 = mcgsm.loglikelihood(inputs, outputs)
loglik1 = []
N = 1000
# estimate log-likelihood via sampling
for _ in range(N):
labels = mcgsm.sample_prior(inputs)
loglik1.append(mcgsm.loglikelihood(inputs, outputs, labels))
loglik1 = vstack(loglik1)
d = abs(logmeanexp(loglik1, 0) - loglik0).ravel()
s = std(loglik1, 0, ddof=1).ravel()
for i in range(M):
self.assertLess(d[i], 6.0 * s[i] / sqrt(N))
def test_sample_conditionally(self):
mcgsm = MCGSM(3, 2, 2, 2, 4)
# make sure there are differences between components
mcgsm.weights = -log(rand(*mcgsm.weights.shape)) * 10.
mcgsm.scales = square(mcgsm.scales * 3.)
inputs = randn(mcgsm.dim_in, 100000)
# sample directly
outputs0 = mcgsm.sample(inputs)
# sample indirectly
labels = mcgsm.sample_prior(inputs)
outputs1 = mcgsm.sample(inputs, labels)
p = ks_2samp(outputs0.ravel(), outputs1.ravel())[1]
self.assertGreater(p, 1e-5)
def test_sample_conditionally(self):
mcgsm = MCGSM(3, 2, 2, 2, 4)
# make sure there are differences between components
mcgsm.weights = -log(rand(*mcgsm.weights.shape)) * 10.0
mcgsm.scales = square(mcgsm.scales * 3.0)
inputs = randn(mcgsm.dim_in, 100000)
# sample directly
outputs0 = mcgsm.sample(inputs)
# sample indirectly
labels = mcgsm.sample_prior(inputs)
outputs1 = mcgsm.sample(inputs, labels)
p = ks_2samp(outputs0.ravel(), outputs1.ravel())[1]
self.assertGreater(p, 1e-5)
