# num_dstates = np.prod([rv.dstates for rv in cond_g.Vd])
if num_dstates > 1000:
print(
'too many dstates, exact mode finding might take a while, consider parallelizing...'
)
for i, rv in enumerate(query_rvs):
m = get_rv_marg_map_from_bn_params(*bn, cond_g.Vd_idx,
cond_g.Vc_idx, rv)
mmap[i] = m
assert rv.domain_type[
0] == 'c', 'only looking at kl for cnode queries for now'
crv_idx = cond_g.Vc_idx[rv]
crv_marg_params = get_crv_marg(*bn, crv_idx)
marg_logpdf = utils.get_scalar_gm_log_prob(
None, *crv_marg_params, get_fun=True)
margs[i] = marg_logpdf
if baseline == 'gibbs':
num_burnin = 200
num_samples = 1000
num_gm_components_for_crv = 3
disc_block_its = 40
hgsampler = HybridGaussianSampler(converted_factors, cond_g.Vd,
cond_g.Vc, cond_g.Vd_idx,
cond_g.Vc_idx)
hgsampler.block_gibbs_sample(num_burnin=num_burnin,
num_samples=num_samples,
disc_block_its=disc_block_its)
# np.save('cont_samples', hgsampler.cont_samples)
# np.save('disc_samples', hgsampler.disc_samples)
# test_crv_idx = 1
test_crv_marg_params = get_crv_marg(*bn, test_crv_idx)
print(f'true crv{test_crv_idx} marg params', test_crv_marg_params)
osi_test_crv_marg_params = osi.params['w'], osi.params['Mu'][
test_crv_idx], osi.params['Var'][test_crv_idx]
print(f'osi crv{test_crv_idx} marg params', osi_test_crv_marg_params)
import matplotlib.pyplot as plt
plt.figure()
xs = np.linspace(-10, 10, 100)
plt.hist(cont_samples[:, test_crv_idx], normed=True, label='samples')
plt.plot(xs,
np.exp(
utils.get_scalar_gm_log_prob(xs,
w=test_crv_marg_params[0],
mu=test_crv_marg_params[1],
var=test_crv_marg_params[2])),
label='ground truth marg pdf')
plt.plot(xs,
np.exp(
utils.get_scalar_gm_log_prob(xs,
w=osi_test_crv_marg_params[0],
mu=osi_test_crv_marg_params[1],
var=osi_test_crv_marg_params[2])),
label='OSI marg pdf')
plt.legend(loc='best')
# plt.show()
save_name = __file__.split('.py')[0]
plt.savefig('%s.png' % save_name)