def perform_GHC(self):
current_solution = self._bayesian_network
self._best_score = current_score = self._get_score(current_solution)
# draw(self._bayesian_network)
# plt.show()
print 'Initial score :', self._best_score
undirected_graph = current_solution.to_undirected()
change_count = 0
max_count = self._max_change_count
print max_count
while True:
# Pick a random edge and decide the best action to be
# applied on the edge
random_edge = GraphUtils.get_random_edge(self._node_names)
# print random_edge, ' is the edge selected'
current_score, current_solution = \
self._get_best_local_solution(current_solution,
undirected_graph, random_edge)
undirected_graph = current_solution.to_undirected()
if current_score > self._best_score:
change_count = 0
# Update the new best solution
self._best_solution = deepcopy(current_solution)
self._best_score = current_score
print '-----------', self._best_score , '------------------'
else:
change_count += 1
self._add_solution_to_tabu_list(current_solution)
if change_count == max_count:
break
def get_bayesian_networks(self):
bayesian_network = GraphUtils.read_graph(self._network_name + '-' + str(self._max_parents))
bayesian_networks = []
if bayesian_network == None:
# Generate a tree (graph) with required number of nodes.
bayesian_network = self._generate_initial_graph()
bayesian_network.name = self._network_name + '-' + str(self._max_parents)
GraphUtils.write_graph(bayesian_network)
# theoretical bound is infinity but this also does well
num_iterations = 4 * self._num_nodes * self._num_nodes
# Since connectedness is only defined for undirected graphs
# so we have to keep a copy of the bayesian network
# except that all the edges are undirected
undirected_BN = bayesian_network.to_undirected()
bayesian_network.name = self._network_name
# Repeat for a large number of times.
for i in xrange(self._num_BNs):
count, i, j = 0, 0, 0;
while count < num_iterations:
edge = (i, j) = GraphUtils.get_random_edge(self._node_names)
if bayesian_network.has_edge(*edge):
# If (i,j) is in the graph, remove it
GraphUtils.apply_action(bayesian_network, undirected_BN, edge, 'remove', self._max_parents)
else:
# If the edge (i,j) is not in the graph, add it.
GraphUtils.apply_action(bayesian_network, undirected_BN, edge, 'add', self._max_parents)
count += 1
bayesian_networks.append(deepcopy(bayesian_network))
# Return the obtained graph
return bayesian_networks
