def optimize_local_meanfield(global_natparam, node_potentials, tol=1e-3, max_iter=100):
label_global, gaussian_global = global_natparam
local_vlb = -np.inf
label_stats = initialize_local_meanfield(label_global, node_potentials)
for _ in xrange(max_iter):
gaussian_natparam, gaussian_stats, gaussian_vlb = \
gaussian_meanfield(gaussian_global, unbox(node_potentials), label_stats)
label_natparam, label_stats, label_vlb = \
label_meanfield(label_global, gaussian_global, gaussian_stats)
local_vlb, prev_local_vlb = label_vlb + gaussian_vlb, local_vlb
if abs(local_vlb - prev_local_vlb) < tol:
break
else:
print 'iteration limit reached'
# recompute values that depend on node_potentials at optimum
gaussian_natparam, gaussian_stats, gaussian_vlb = \
gaussian_meanfield(gaussian_global, node_potentials, label_stats)
label_natparam, label_stats, label_vlb = \
label_meanfield(label_global, gaussian_global, gaussian_stats)
stats = label_stats, gaussian_stats
local_natparams = label_natparam, gaussian_natparam
vlbs = label_vlb, gaussian_vlb
return stats, local_natparams, vlbs
def run_inference(prior_natparam, global_natparam, nn_potentials, num_samples):
hmm_global_natparam, lds_global_natparam = global_natparam
# optimize local mean field (can use unboxed val for low-level code)
(hmm_stats, _), (hmm_local_natparam, lds_local_natparam), _ = \
optimize_local_meanfield(global_natparam, unbox(nn_potentials))
# recompute terms that depend on nn_potentials at optimum (using boxed val)
samples, lds_stats, lds_normalizer = natural_lds_inference_general(
lds_local_natparam, nn_potentials, num_samples)
hmm_vlb = get_hmm_vlb(lds_global_natparam, hmm_local_natparam, lds_stats)
# get global statistics from the local expected stats
global_lds_stats, local_lds_stats = lds_stats[:-1], lds_stats[-1]
expected_stats = get_global_stats(hmm_stats, global_lds_stats)
# compute global and local vlb terms
global_vlb = slds_prior_vlb(global_natparam, prior_natparam)
lds_vlb = lds_normalizer - contract(nn_potentials, local_lds_stats)
local_vlb = hmm_vlb + lds_vlb
return samples, expected_stats, global_vlb, local_vlb
def run_inference(prior_natparam, global_natparam, nn_potentials, num_samples):
_, stats, local_natparam, local_kl = local_meanfield(global_natparam, nn_potentials)
samples = gaussian.natural_sample(local_natparam[1], num_samples)
global_kl = prior_kl(global_natparam, prior_natparam)
return samples, unbox(stats), global_kl, local_kl
def run_inference(prior_natparam, global_natparam, nn_potentials, num_samples):
_, stats, local_natparam, local_kl = local_meanfield(global_natparam, nn_potentials)
samples = gaussian.natural_sample(local_natparam[1], num_samples)
global_kl = prior_kl(global_natparam, prior_natparam)
return samples, unbox(stats), global_kl, local_kl
def run_inference(prior_natparam, global_natparam, nn_potentials):
stats, local_natparam, local_kl = local_meanfield(global_natparam,
nn_potentials)
global_kl = prior_kl(global_natparam, prior_natparam)
return unbox(stats), global_kl, local_kl