def run_bayes_bp(self, fold=10):
nodes_with_in_degree = []
for node in self.mybp.G.nodes():
nodes_with_in_degree.append([node, self.mybp.G.in_degree(node)])
nodes = sorted(nodes_with_in_degree, key=lambda x: x[1], reverse=True)
trunk_size = len(nodes) / fold
train_ids = [node[0] for node in nodes[0:trunk_size]]
test_ids = [node[0] for node in nodes[trunk_size:]]
fscore = 0.0
accuracy = 0.0
"""apply Belief Propagation"""
nodes_belief = dict.fromkeys(self.mybp.G.nodes(), [0.5, 0.5])
for node in train_ids:
if self.mybp.nodes_label[node] == "d":
nodes_belief[node] = [0.99, 0.01]
else:
nodes_belief[node] = [0.01, 0.99]
edge_messages = self.mybp.initialize_messages(self.mybp.G)
new_nodes_belief, edge_messages = bp.beliefprop(
self.mybp.G, edge_messages, nodes_belief)
accuracy, fscore = self.mybp.evaluation(self.mybp.nodes_label,
new_nodes_belief, train_ids)
print "\tbayes_based_bp: ", accuracy, fscore
return accuracy, fscore
def run_bayes_bp(self, fold = 10):
nodes_with_in_degree = []
for node in self.mybp.G.nodes():
nodes_with_in_degree.append([node, self.mybp.G.in_degree(node)])
nodes = sorted(nodes_with_in_degree, key = lambda x : x[1], reverse = True)
trunk_size = len(nodes)/fold
train_ids = [node[0] for node in nodes[0:trunk_size]]
test_ids = [node[0] for node in nodes[trunk_size:]]
fscore = 0.0
accuracy = 0.0
"""apply Belief Propagation"""
nodes_belief = dict.fromkeys(self.mybp.G.nodes(), [0.5, 0.5])
for node in train_ids:
if self.mybp.nodes_label[node] == "d":
nodes_belief[node] = [0.99, 0.01]
else:
nodes_belief[node] = [0.01, 0.99]
edge_messages = self.mybp.initialize_messages(self.mybp.G)
new_nodes_belief, edge_messages = bp.beliefprop(self.mybp.G, edge_messages, nodes_belief)
accuracy, fscore = self.mybp.evaluation(self.mybp.nodes_label, new_nodes_belief, train_ids)
print "\tbayes_based_bp: ", accuracy, fscore
return accuracy, fscore
def run_bayes_bp(self, fold=10, iters=5):
nodes = list(self.mybp.G.nodes())
f = open(
"../results/results_fscore_accuracy_feb27_bp/bp_result_bayes_adjusted_%dfold.csv"
% fold, "w")
f.write("bayes_based_bp,bayes_adjusted_bp\n")
f.write(
"iter,bayes_adjusted_bp_fscore,bayes_adjusted_bp_accuracy,bayes_fscore,bayes_accuracy\n"
)
for iter in range(iters):
random.shuffle(nodes) # shuffle the list for every iteration
trunk_size = len(nodes) / fold
mean_fscore_bayes = 0.0
mean_accuracy_bayes = 0.0
mean_fscore_bayes_adjusted_bp = 0.0
mean_accuracy_bayes_adjusted_bp = 0.0
for i in range(fold):
print "iter %d " % i
train_ids = []
test_ids = []
for j in range(fold):
if j == i:
if j == fold - 1:
train_ids.extend(nodes[j * trunk_size:len(nodes)])
else:
train_ids.extend(nodes[j * trunk_size:(j + 1) *
trunk_size])
else:
if j == fold - 1:
test_ids.extend(nodes[j * trunk_size:len(nodes)])
else:
test_ids.extend(nodes[j * trunk_size:(j + 1) *
trunk_size])
"""apply Naive Bayes"""
train_x = []
train_Y = []
test_x = []
test_Y = []
for id in train_ids:
train_x.append(self.toList(self.vector[id]))
if self.mybp.nodes_label[id] == "d":
train_Y.append(1)
else:
train_Y.append(0)
for id in test_ids:
test_x.append(self.toList(self.vector[id]))
if self.mybp.nodes_label[id] == "d":
test_Y.append(1)
else:
test_Y.append(0)
clf = GaussianNB()
clf.fit(train_x, train_Y)
y_predicted = clf.predict(test_x)
y_predicted_prob = clf.predict_proba(test_x)
print y_predicted_prob
f_score_bayes = metrics.f1_score(test_Y, y_predicted)
accuracy_bayes = metrics.accuracy_score(test_Y, y_predicted)
mean_fscore_bayes += f_score_bayes
mean_accuracy_bayes += accuracy_bayes
print "\tnaive bayes: ", accuracy_bayes, f_score_bayes
"""apply bayes_adjusted Belief Propagation"""
nodes_belief = dict.fromkeys(self.mybp.G.nodes(), [0.5, 0.5])
for node in train_ids:
if self.mybp.nodes_label[node] == "d":
nodes_belief[node] = [0.99, 0.01]
else:
nodes_belief[node] = [0.01, 0.99]
edge_messages = self.mybp.initialize_messages(self.mybp.G)
new_nodes_belief, edge_messages = bp.beliefprop(
self.mybp.G, edge_messages, nodes_belief)
predicted_prob = {}
for k in range(len(test_ids)):
predicted_prob.update({test_ids[k]: y_predicted_prob[k]})
accuracy_bayes_adjusted_bp, fscore_bayes_adjusted_bp = self.evaluate(
self.mybp.nodes_label, new_nodes_belief, train_ids,
predicted_prob)
mean_accuracy_bayes_adjusted_bp += accuracy_bayes_adjusted_bp
mean_fscore_bayes_adjusted_bp += fscore_bayes_adjusted_bp
print "\tbayes_adjusted_bp: ", accuracy_bayes_adjusted_bp, fscore_bayes_adjusted_bp
f.write(
"f%d,%f,%f,%f,%f\n" %
(i, fscore_bayes_adjusted_bp, accuracy_bayes_adjusted_bp,
f_score_bayes, accuracy_bayes))
f.write("i%d,%f,%f,%f,%f\n" %
(iter, mean_fscore_bayes_adjusted_bp / float(fold),
mean_accuracy_bayes_adjusted_bp / float(fold),
mean_fscore_bayes / float(fold),
mean_accuracy_bayes / float(fold)))
f.close()
def run_bayes_bp(self, fold = 10, iters = 5):
nodes = list(self.mybp.G.nodes())
f = open("../results/results_fscore_accuracy_feb27_bp/bp_result_bayes_adjusted_%dfold.csv" %fold,"w")
f.write("bayes_based_bp,bayes_adjusted_bp\n")
f.write("iter,bayes_adjusted_bp_fscore,bayes_adjusted_bp_accuracy,bayes_fscore,bayes_accuracy\n")
for iter in range(iters):
random.shuffle(nodes) # shuffle the list for every iteration
trunk_size = len(nodes)/fold
mean_fscore_bayes = 0.0
mean_accuracy_bayes = 0.0
mean_fscore_bayes_adjusted_bp = 0.0
mean_accuracy_bayes_adjusted_bp = 0.0
for i in range(fold):
print "iter %d " %i
train_ids = []
test_ids = []
for j in range(fold):
if j == i:
if j == fold-1:
train_ids.extend(nodes[j*trunk_size:len(nodes)])
else:
train_ids.extend(nodes[j*trunk_size:(j+1)*trunk_size])
else:
if j == fold-1:
test_ids.extend(nodes[j*trunk_size:len(nodes)])
else:
test_ids.extend(nodes[j*trunk_size:(j+1)*trunk_size])
"""apply Naive Bayes"""
train_x = []
train_Y = []
test_x = []
test_Y = []
for id in train_ids:
train_x.append(self.toList(self.vector[id]))
if self.mybp.nodes_label[id] == "d":
train_Y.append(1)
else:
train_Y.append(0)
for id in test_ids:
test_x.append(self.toList(self.vector[id]))
if self.mybp.nodes_label[id] == "d":
test_Y.append(1)
else:
test_Y.append(0)
clf = GaussianNB()
clf.fit(train_x, train_Y)
y_predicted = clf.predict(test_x)
y_predicted_prob = clf.predict_proba(test_x)
print y_predicted_prob
f_score_bayes = metrics.f1_score(test_Y, y_predicted)
accuracy_bayes = metrics.accuracy_score(test_Y, y_predicted)
mean_fscore_bayes += f_score_bayes
mean_accuracy_bayes += accuracy_bayes
print "\tnaive bayes: ", accuracy_bayes, f_score_bayes
"""apply bayes_adjusted Belief Propagation"""
nodes_belief = dict.fromkeys(self.mybp.G.nodes(), [0.5, 0.5])
for node in train_ids:
if self.mybp.nodes_label[node] == "d":
nodes_belief[node] = [0.99, 0.01]
else:
nodes_belief[node] = [0.01, 0.99]
edge_messages = self.mybp.initialize_messages(self.mybp.G)
new_nodes_belief, edge_messages = bp.beliefprop(self.mybp.G, edge_messages, nodes_belief)
predicted_prob = {}
for k in range(len(test_ids)):
predicted_prob.update({test_ids[k] : y_predicted_prob[k]})
accuracy_bayes_adjusted_bp, fscore_bayes_adjusted_bp = self.evaluate(self.mybp.nodes_label,
new_nodes_belief,
train_ids,
predicted_prob)
mean_accuracy_bayes_adjusted_bp += accuracy_bayes_adjusted_bp
mean_fscore_bayes_adjusted_bp += fscore_bayes_adjusted_bp
print "\tbayes_adjusted_bp: ", accuracy_bayes_adjusted_bp, fscore_bayes_adjusted_bp
f.write("f%d,%f,%f,%f,%f\n" %(i, fscore_bayes_adjusted_bp, accuracy_bayes_adjusted_bp,
f_score_bayes, accuracy_bayes))
f.write("i%d,%f,%f,%f,%f\n" %(iter, mean_fscore_bayes_adjusted_bp/float(fold),
mean_accuracy_bayes_adjusted_bp/float(fold),
mean_fscore_bayes/float(fold),
mean_accuracy_bayes/float(fold)))
f.close()