Fixing the _categorical_exact function and the tests to be more deter…

…ministic (#1059)

Thanks so much for finding this issue and submitting a great PR fixing it.

pomegranate/bayesian_network.py

@@ -954,8 +954,8 @@ def _categorical_exact(X, sample_weight=None, include_parents=None,
 				order_graph.add_edge(parent, subset, weight=weight,
 					structure=structure)
 
-	path = nx.shortest_path(order_graph, source=(), target=tuple(range(d)),
-		weight='weight')
+	path = sorted(nx.all_shortest_paths(order_graph, source=(),
+		target=tuple(range(d)), weight="weight"))[1]
 
 	score, structure = 0, list( None for i in range(d) )
 	for u, v in zip(path[:-1], path[1:]):

tests/test_bayesian_network.py

@@ -1030,33 +1030,33 @@ def test_learn_structure_exact(X):
 	model = BayesianNetwork(algorithm='exact')
 	model.fit(X)
 
-	assert_tuple_equal(model._parents, ((), (), (), (1, 2)))
+	assert_tuple_equal(model._parents, ((), (0, 2), (), ()))
 
 	assert_array_almost_equal(model.distributions[0].probs,
 		[[0.4545, 0.5455]], 4)
-	assert_array_almost_equal(model.distributions[1].probs, 
+	assert_array_almost_equal(model.distributions[3].probs, 
 		[[0.4545, 0.5455]], 4)
 	assert_array_almost_equal(model.distributions[2].probs,
 		[[0.5455, 0.4545]], 4)
-	assert_array_almost_equal(model.distributions[3].probs[0], 
-		[[[1., 0.],
-          [1., 0.]],
+	assert_array_almost_equal(model.distributions[1].probs[0], 
+		[[[0.3333, 0.6667],
+          [1.0000, 0.0000]],
 
-         [[0., 1.],
-          [0., 1.]]], 4)
+         [[0.3333, 0.6667],
+          [0.3333, 0.6667]]], 4)
 
 	assert_array_almost_equal(model._factor_graph.factors[0].probs, 
 		[[0.4545, 0.5455]], 4)
-	assert_array_almost_equal(model._factor_graph.factors[1].probs, 
+	assert_array_almost_equal(model._factor_graph.factors[3].probs, 
 		[[0.4545, 0.5455]], 4)
 	assert_array_almost_equal(model._factor_graph.factors[2].probs, 
 		[[0.5455, 0.4545]], 4)
-	assert_array_almost_equal(model._factor_graph.factors[3].probs, 
-		[[[0.25, 0.],
-          [0.25, 0.]],
+	assert_array_almost_equal(model._factor_graph.factors[1].probs, 
+		[[[0.0833, 0.1667],
+          [0.2500, 0.0000]],
 
-         [[0., 0.25],
-          [0., 0.25]]], 4)
+         [[0.0833, 0.1667],
+          [0.0833, 0.1667]]], 4)
 
 
 def test_summarize(X, distributions):

tests/test_bayesian_network_structure_learning.py

@@ -152,18 +152,18 @@ def test_categorical_chow_liu_raises(X, w):
 
 def test_categorical_exact(X):
 	structure = _categorical_exact(X)
-	assert_tuple_equal(structure, ((), (2,), (), (1, 2)))
+	assert_tuple_equal(structure, ((), (0,), (), (0, 1)))
 
 	structure = _categorical_exact(X, max_parents=1)
-	assert_tuple_equal(structure, ((), (2,), (), ()))
+	assert_tuple_equal(structure, ((), (0,), (), ()))
 
 
 def test_categorical_exact_weighted(X, w):
 	structure = _categorical_exact(X, w)
-	assert_tuple_equal(structure, ((), (2,), (), (1, 2)))
+	assert_tuple_equal(structure, ((), (0,), (), (0, 1)))
 
 	structure = _categorical_exact(X, w, max_parents=1)
-	assert_tuple_equal(structure, ((), (2,), (), ()))
+	assert_tuple_equal(structure, ((), (0,), (), ()))
 
 
 def test_categorical_exact_exclude_parents(X):
@@ -173,35 +173,35 @@ def test_categorical_exact_exclude_parents(X):
 
 	structure = _categorical_exact(X, exclude_parents=exclude_parents, 
 		max_parents=1)
-	assert_tuple_equal(structure, ((), (), (), (0,)))
+	assert_tuple_equal(structure, ((), (), (0,), ()))
 
 	exclude_parents = ((), (2,), (), (0, 1))
 	structure = _categorical_exact(X, exclude_parents=exclude_parents)
-	assert_tuple_equal(structure, ((2, 3), (), (), (2,)))
+	assert_tuple_equal(structure, ((3,), (), (0,3), ()))
 
 
 def test_categorical_exact_large():
 	numpy.random.seed(0)
 	X = numpy.random.randint(3, size=(50, 10))
 
 	structure = _categorical_exact(X)
-	assert_tuple_equal(structure, ((), (6, 7), (), (1, 6, 7), (), (1, 3, 6, 7), 
-		(), (6,), (), ()))
+	assert_tuple_equal(structure, ((1, 2), (), (), (), (0, 1, 2), (), (), 
+		(), (0, 1, 2, 4), ()))
 
 	structure = _categorical_exact(X, max_parents=1)
-	assert_tuple_equal(structure, ((), (), (), (), (), (), (), (3,), (), ()))
+	assert_tuple_equal(structure, ((), (), (), (), (), (), (), (2,), (), ()))
 
 	structure = _categorical_exact(X, max_parents=2)
-	assert_tuple_equal(structure, ((), (), (6, 7), (), (), (), (), (6,), (), 
+	assert_tuple_equal(structure, ((), (), (), (), (), (), (), (0, 1), (), 
 		()))
 
 	structure = _categorical_exact(X, max_parents=3)
-	assert_tuple_equal(structure, ((), (), (), (), (), (6, 7), (), (6,), 
-		(5, 6, 7), ()))
+	assert_tuple_equal(structure, ((1, 2), (), (), (), (0, 1, 2), (), (), (), 
+		(), ()))
 
 	structure = _categorical_exact(X, max_parents=4)
-	assert_tuple_equal(structure, ((), (6, 7), (), (1, 6, 7), (), (1, 3, 6, 7), 
-		(), (6,), (), ()))
+	assert_tuple_equal(structure, ((1, 2), (), (), (), (0, 1, 2), (), (), (), 
+		(0, 1, 2, 4), ()))
 
 
 def test_categorical_exact_large_pseudocount():
@@ -224,7 +224,7 @@ def test_categorical_exact_large_pseudocount():
 	assert_tuple_equal(structure, ((), (), (), (), (), (), (), ()))
 
 	structure = _categorical_exact(X, pseudocount=1e-8)
-	assert_tuple_equal(structure, ((5,), (), (), (), (), (), (), ()))
+	assert_tuple_equal(structure, ((), (), (), (), (), (0,), (), ()))
 
 
 ###