def test_mcmc(self):
nlevels = 5
self.create_base_dataset(
sample_size=1000,
num_numeric=3,
num_cat=4,
num_levels=nlevels)
ac = R.AutoClean()
niter = 100
nclusters = 7
ac.train_model(self._data, nclusters=nclusters, niter=niter)
beta = self._beta
xdim = beta.shape[0]
ydim = beta.shape[1]
self.assertEqual(ac.coefficients.shape, (niter, xdim, ydim))
self.assertEqual(ac.residual_variance.shape, (niter, ydim, ydim))
self.assertEqual(len(ac.atom_probs), ydim)
self.assertEqual(ac.atom_probs["X1"].shape,
(niter, nclusters, 1))
self.assertTrue(
np.allclose(ac.atom_probs["X1"].sum(axis=2),
np.ones((niter, nclusters)))
)
self.assertEqual(len(ac.atom_error_probs), ydim)
self.assertEqual(ac.atom_error_probs["X1"].shape,
(niter, nclusters, 1, 2))
self.assertTrue(np.allclose(
ac.atom_error_probs["X1"].sum(axis=3),
np.ones((niter, nclusters, 1))))
self.assertEqual(len(ac.level_probs), self._ncat)
self.assertEqual(ac.level_probs["cat1"].shape,
(niter, nclusters, nlevels))
self.assertTrue(np.allclose(
ac.level_probs["cat1"].sum(axis=2),
np.ones((niter, nclusters))))
self.assertEqual(len(ac.level_observation_probs), self._ncat)
self.assertEqual(ac.level_observation_probs["cat1"].shape,
(niter, nclusters, nlevels, nlevels+1))
self.assertTrue(np.allclose(
ac.level_observation_probs["cat1"].sum(axis=3),
np.ones((niter, nclusters, nlevels))))
to_impute = self._data.iloc[[3, 9, 24], :]
iterations = [50, 60, 70, 75, 90]
imputed = ac.impute_rows(to_impute, iterations)
self.assertEqual(len(imputed), len(iterations))
self.assertEqual(imputed[0].shape, to_impute.shape)
def test_autoclean_construction(self):
ac = R.AutoClean()
ac.train_model(self._data, nclusters=3, niter=10)
