def test_pdf(self, distribution_test):
# 1d case
x = [1, 2, -1]
means = [0, 2]
weights = normalize_weights([.4, .1])
d = GMDistribution.pdf(x, means, weights=weights)
d_true = weights[0] * ss.norm.pdf(x, loc=means[0]) + \
weights[1] * ss.norm.pdf(x, loc=means[1])
assert np.allclose(d, d_true)
# Test with a single observation
# assert GMDistribution.pdf(x[0], means, weights=weights).ndim == 0
# Distribution_test with 1d means
distribution_test(GMDistribution, means, weights=weights)
# 2d case
x = [[1, 2, -1], [0, 0, 2]]
means = [[0, 0, 0], [-1, -.2, .1]]
d = GMDistribution.pdf(x, means, weights=weights)
d_true = weights[0] * ss.multivariate_normal.pdf(x, mean=means[0]) + \
weights[1] * ss.multivariate_normal.pdf(x, mean=means[1])
assert np.allclose(d, d_true)
# Test with a single observation
assert GMDistribution.pdf(x[0], means, weights=weights).ndim == 0
# Distribution_test with 3d means
distribution_test(GMDistribution, means, weights=weights)
def _compute_weights_and_cov(self, pop):
params = np.column_stack(tuple([pop.outputs[p] for p in self.parameter_names]))
if self._populations:
q_logpdf = GMDistribution.logpdf(params, *self._gm_params)
p_logpdf = self._prior.logpdf(params)
w = np.exp(p_logpdf - q_logpdf)
else:
w = np.ones(pop.n_samples)
if np.count_nonzero(w) == 0:
raise RuntimeError("All sample weights are zero. If you are using a prior "
"with a bounded support, this may be caused by specifying "
"a too small sample size.")
# New covariance
cov = 2 * np.diag(weighted_var(params, w))
if not np.all(np.isfinite(cov)):
logger.warning("Could not estimate the sample covariance. This is often "
"caused by majority of the sample weights becoming zero."
"Falling back to using unit covariance.")
cov = np.diag(np.ones(params.shape[1]))
return w, cov
def prepare_new_batch(self, batch_index):
"""Prepare values for a new batch.
Parameters
----------
batch_index : int
next batch_index to be submitted
Returns
-------
batch : dict or None
Keys should match to node names in the model. These values will override any
default values or operations in those nodes.
"""
if self.state['round'] == 0:
# Use the actual prior
return
# Sample from the proposal, condition on actual prior
params = GMDistribution.rvs(*self._gm_params, size=self.batch_size,
prior_logpdf=self._prior.logpdf,
random_state=self._round_random_state)
batch = arr2d_to_batch(params, self.parameter_names)
return batch
def test_rvs_prior_ok(self):
means = [0.8, 0.5]
weights = [.3, .7]
N = 10000
prior_logpdf = ss.uniform(0, 1).logpdf
rvs = GMDistribution.rvs(means, weights=weights, size=N, prior_logpdf=prior_logpdf)
# Ensure prior pdf > 0 for all samples
assert np.all(np.isfinite(prior_logpdf(rvs)))
def prepare_new_batch(self, batch_index):
if self.state['round'] == 0:
# Use the actual prior
return
# Sample from the proposal
params = GMDistribution.rvs(*self._gm_params, size=self.batch_size,
random_state=self._round_random_state)
batch = arr2d_to_batch(params, self.parameter_names)
return batch
def test_rvs(self):
means = [[1000, 3], [-1000, -3]]
weights = [.3, .7]
N = 10000
random = np.random.RandomState(12042017)
rvs = GMDistribution.rvs(means, weights=weights, size=N, random_state=random)
rvs = rvs[rvs[:, 0] < 0, :]
# Test correct proportion of samples near the second mode
assert np.abs(len(rvs) / N - .7) < .01
# Test that the mean of the second mode is correct
assert np.abs(np.mean(rvs[:, 1]) + 3) < .1