def inference_ad3(unary_potentials,
pairwise_potentials,
edges,
relaxed=False,
verbose=0,
return_energy=False):
import AD3
shape_org = unary_potentials.shape[:-1]
n_states, pairwise_potentials = \
_validate_params(unary_potentials, pairwise_potentials, edges)
unaries = unary_potentials.reshape(-1, n_states)
res = AD3.general_graph(unaries,
edges,
pairwise_potentials,
verbose=verbose)
unary_marginals, pairwise_marginals, energy = res
#n_fractional = np.sum(unary_marginals.max(axis=-1) < .99)
#if n_fractional:
#print("fractional solutions found: %d" % n_fractional)
if relaxed:
unary_marginals = unary_marginals.reshape(unary_potentials.shape)
y = (unary_marginals, pairwise_marginals)
else:
y = np.argmax(unary_marginals, axis=-1)
y = y.reshape(shape_org)
if return_energy:
return y, -energy
return y
def _inference_ad3(x, unary_params, pairwise_params, edges,
relaxed=False, verbose=0):
raise NotImplementedError("AD3 doesn't work on graphs yet!")
res = AD3.simple_grid(unary_params * x, pairwise_params, verbose=verbose)
unary_marginals, pairwise_marginals, energy = res
n_fractional = np.sum(unary_marginals.max(axis=-1) < .99)
if n_fractional:
print("fractional solutions found: %d" % n_fractional)
if relaxed:
unary_marginals = unary_marginals.reshape(x.shape)
pairwise_accumulated = pairwise_marginals.sum(axis=0)
pairwise_accumulated = pairwise_accumulated.reshape(x.shape[-1],
x.shape[-1])
y = (unary_marginals, pairwise_accumulated)
else:
y = np.argmax(unary_marginals, axis=-1)
y = y.reshape(x.shape[0], x.shape[1])
return y
def inference_ad3(unary_potentials, pairwise_potentials, edges, relaxed=False,
verbose=0, return_energy=False):
import AD3
shape_org = unary_potentials.shape[:-1]
n_states, pairwise_potentials = \
_validate_params(unary_potentials, pairwise_potentials, edges)
unaries = unary_potentials.reshape(-1, n_states)
res = AD3.general_graph(unaries, edges, pairwise_potentials,
verbose=verbose)
unary_marginals, pairwise_marginals, energy = res
#n_fractional = np.sum(unary_marginals.max(axis=-1) < .99)
#if n_fractional:
#print("fractional solutions found: %d" % n_fractional)
if relaxed:
unary_marginals = unary_marginals.reshape(unary_potentials.shape)
y = (unary_marginals, pairwise_marginals)
else:
y = np.argmax(unary_marginals, axis=-1)
y = y.reshape(shape_org)
if return_energy:
return y, -energy
return y