def testFortran(self):
"""Test if `normalize_logspace` works with a F-contiguous array"""
np.random.seed(0)
mat = random((NCOL, NROW)).T
self.assertTrue(mat.flags["F_CONTIGUOUS"])
mat_out = normalize_logspace(mat)
row_sum = mat_out.sum(1)
approx_equal = arrays_almost_equal(row_sum, np.ones(NROW), accuracy=ACC)
self.assertTrue(approx_equal)
def _update_latent_resp(data, smm_dof, posterior_nws_scale,
posterior_nws_dof, log_smm_mixweight,
log_det_precision, scatter):
""" Update `latent_resp` (Eq 22 in Arch2007) """
num_features = data.shape[1]
latent_resp = (gammaln((num_features + smm_dof) / 2) -
gammaln(smm_dof / 2) -
(num_features / 2) * np.log(smm_dof * pi) +
log_smm_mixweight + log_det_precision / 2 -
((num_features + smm_dof) / 2) *
np.log(1 +
(posterior_nws_dof / smm_dof).T * scatter.T +
(num_features /
(smm_dof * posterior_nws_scale)).T))
latent_resp = normalize_logspace(latent_resp)
return latent_resp
def _update_latent_resp(data, smm_dof, posterior_nws_scale,
posterior_nws_dof, log_smm_mixweight,
log_det_precision, scatter):
""" Update `latent_resp` (Eq 22 in Arch2007) """
num_features = data.shape[1]
num_comp = len(log_smm_mixweight)
def get_exp_latent(k):
return (gammaln((num_features + smm_dof[k]) / 2) -
gammaln(smm_dof[k] / 2) -
(num_features / 2) * log(smm_dof[k] * pi) +
log_smm_mixweight[k] + log_det_precision[k] / 2 -
((num_features + smm_dof[k]) / 2) *
log(1 + (posterior_nws_dof[k] / smm_dof[k]) *
scatter[k, :] +
num_features / (smm_dof[k] * posterior_nws_scale[k])))
latent_resp = np.array([get_exp_latent(k)
for k in range(num_comp)]).T
latent_resp = normalize_logspace(latent_resp)
return latent_resp
def init_d2_weighting(data, num_comp):
num_obs = data.shape[0]
cov_inv = np.linalg.inv(np.cov(data, rowvar=0))
select_prob = np.ones(num_obs) / num_obs
shortest_dist = np.inf * np.ones(num_obs)
centroid = np.ones(num_comp)
for k in range(num_comp):
# Select a random data point as centroid
centroid[k] = np.nonzero(multinomial(1, select_prob))[0]
# Recompute distances
for i, d in enumerate(shortest_dist):
d_new = mahalanobis(data[centroid[k], :], data[i, :], cov_inv)
if d_new < d: shortest_dist[i] = d_new
select_prob = normalize_logspace(
pow(shortest_dist.reshape(1, len(shortest_dist)), 2, 1))
select_prob = select_prob.flatten()
return centroid
