def _sample_adjacency_matrix_without_loose_ends(self, adjacency_matrix, node):
req_num_parents = self.num_parents_per_node[str(node)]
current_num_parents = np.sum(adjacency_matrix[:, node], dtype=np.int)
num_parents_left = req_num_parents - current_num_parents
sampled_parents = \
random.sample(list(parent_combinations_old(adjacency_matrix, node, n_parents=num_parents_left)), 1)[0]
for parent in sampled_parents:
adjacency_matrix[parent, node] = 1
adjacency_matrix = self._sample_adjacency_matrix_without_loose_ends(adjacency_matrix, parent)
return adjacency_matrix
def _generate_adjacency_matrix(self, adjacency_matrix, node):
if self._check_validity_of_adjacency_matrix(adjacency_matrix):
# If graph from search space then yield.
yield adjacency_matrix
else:
req_num_parents = self.num_parents_per_node[str(node)]
current_num_parents = np.sum(adjacency_matrix[:, node], dtype=np.int)
num_parents_left = req_num_parents - current_num_parents
for parents in parent_combinations_old(adjacency_matrix, node, n_parents=num_parents_left):
# Make copy of adjacency matrix so that when it returns to this stack
# it can continue with the unmodified adjacency matrix
adjacency_matrix_copy = copy.copy(adjacency_matrix)
for parent in parents:
adjacency_matrix_copy[parent, node] = 1
for graph in self._generate_adjacency_matrix(adjacency_matrix=adjacency_matrix_copy, node=parent):
yield graph