def setUp(self): # two separate points, verify that the likelihoods are correct # and the sampling is sound self.points = dict() self.points[1] = Counter() self.points[1]['x'] = 1.0 self.points[1]['y'] = 1.0 self.points[2] = Counter() self.points[2]['x'] = 10.0 self.points[2]['y'] = 10.0 self.sampler = CRPGibbsSampler(self.points)
def train(self, paths):
pre_counts, post_counts, full_counts = self._gather_colocation_counts(
paths)
full_counts += pre_counts
full_counts += post_counts
# and hand over work to the sampler
print len(full_counts)
sampler = CRPGibbsSampler(full_counts, burn_in_iterations=20)
print len(full_counts)
print sampler._datum_to_cluster
class CRPGibbsSamplerTest(unittest.TestCase): def setUp(self): # two separate points, verify that the likelihoods are correct # and the sampling is sound self.points = dict() self.points[1] = Counter() self.points[1]['x'] = 1.0 self.points[1]['y'] = 1.0 self.points[2] = Counter() self.points[2]['x'] = 10.0 self.points[2]['y'] = 10.0 self.sampler = CRPGibbsSampler(self.points) def test_simple_sample_datum(self): self.sampler._sample_datum = lambda datum: 0 self.sampler.log_likelihood = lambda: 0.0 self.sampler.gibbs(1) self.assertTrue(all(v == 0 for v in self.sampler._datum_to_cluster.itervalues())) self.assertTrue(all(len(c) == 0 for idx, c in self.sampler._cluster_to_datum.iteritems() if idx != 0), self.sampler._cluster_to_datum) self.assertEqual(len(self.sampler._cluster_to_datum[0]), 2) self.sampler.gibbs(1) self.assertEqual(self.sampler._cluster_count[0], 1) self.assertEqual(self.sampler._iteration_likelihoods[0], 0.0) def test_two_cluster_sample_datum(self): self.sampler._sample_datum = lambda datum: 0 if datum == self.points[1] else 1 self.sampler.log_likelihood = lambda: 0.0 self.sampler.gibbs(1) self.assertEqual(self.sampler._datum_to_cluster[1], 0) self.assertEqual(self.sampler._datum_to_cluster[2], 1) self.assertEqual(len(self.sampler._cluster_to_datum[0]), 1) self.assertEqual(len(self.sampler._cluster_to_datum[1]), 1) # invert the sampler self.sampler._sample_datum = lambda datum: 1 if datum == self.points[1] else 0 self.sampler.gibbs(1) self.assertEqual(self.sampler._cluster_count[0], 2) self.assertEqual(self.sampler._iteration_likelihoods[0], 0.0) # Assert they switched clusters self.assertEqual(self.sampler._datum_to_cluster[1], 1) self.assertEqual(self.sampler._datum_to_cluster[2], 0) self.assertEqual(len(self.sampler._cluster_to_datum[0]), 1) self.assertEqual(len(self.sampler._cluster_to_datum[1]), 1) def _test_add_self(self): # Shouldn't be initialized yet... self.assertEqual(self.sampler.log_likelihood(), 0.0) self.sampler._add_datum(1, self.points[1], 1) self.sampler._add_datum(2, self.points[2], 2) self.assertEqual(self.sampler.log_likelihood(), -(12.0625 * 2 + 306.25 * 2)) self.assertEqual(len(self.sampler._cluster_to_datum), 2) self.assertEqual(self.sampler._cluster_to_datum[1], [self.points[1]]) self.assertEqual(self.sampler._cluster_to_datum[2], [self.points[2]]) self.sampler._remove_datum(1, self.points[1]) self.sampler._add_datum(1, self.points[1], 2) self.assertEqual(self.sampler.log_likelihood(), -306.625) self.assertEqual(len([c for c, v in self.sampler._cluster_to_datum.iteritems() if v]), 1) self.assertEqual(sorted(self.sampler._cluster_to_datum[2]), sorted(self.points.itervalues()))
