def _sample_component_categorical(model, col_idx, component_idx):
hypers = model['col_hypers'][col_idx]
if component_idx < len(model['col_suffstats'][col_idx]):
suffstats = model['col_suffstats'][col_idx][component_idx]
else:
k = int(model['col_suffstats'][col_idx][0]['k'])
suffstats = {'n': 0., 'k': k}
return csd.sample(suffstats, hypers)
def _sample_component_categorical(model, col_idx, component_idx):
hypers = model['col_hypers'][col_idx]
if component_idx < len(model['col_suffstats'][col_idx]):
suffstats = model['col_suffstats'][col_idx][component_idx]
else:
k = int(model['col_suffstats'][col_idx][0]['k'])
suffstats = {'n': 0., 'k': k}
return csd.sample(suffstats, hypers)
def test_sample_proportions():
n_samples = 10000
hypers = {'dirichlet_alpha': 1.}
suffstats = {'n': 2, 'k': 2, '0': 1, '1': 1}
samples = csd.sample(suffstats, hypers, n=n_samples)
bins = np.array(np.bincount(samples), dtype=float)
assert len(bins) == 2
bins /= n_samples
assert abs(bins[0] - bins[1]) < .05
suffstats = {'n': 10, 'k': 2, '1': 10}
samples = csd.sample(suffstats, hypers, n=n_samples)
bins = np.array(np.bincount(samples), dtype=float)
assert len(bins) == 2
bins /= n_samples
assert abs(bins[0] - bins[1]) > .8
