def train(self, x_supervised, x_unsupervised, y_supervised):
"""
train the modified Naive bayes classifier using both labelled and
unlabelled data. We use the CountVectorizer vectorizaton method from scikit-learn
positional arguments:!
-- X_supervised: [N_sup, in_features]
-- X_unsupervised: [N_unsup, in_features]
-- y_supervised: [N_sup, out_class]
"""
# clf = GaussianNB()
clf = BernoulliNB()
clf.fit(x_supervised, y_supervised)
predi = clf.predict(x_supervised)
old_likelihood = 1
while self.max_rounds > 0:
self.max_rounds -= 1
# E-step
predi = clf.predict(x_unsupervised)
# M-step
clf.fit(x_unsupervised, predi)
# calculate new total likelihood
predi = clf.predict(x_supervised)
unsupervised_log_matrix = clf._joint_log_likelihood(x_unsupervised)
supervised_log_matrix = clf._joint_log_likelihood(x_supervised)
# print("unsupervised_log_matrix before log", unsupervised_log_matrix)
total_likelihood = self.get_log_likelihood(unsupervised_log_matrix, supervised_log_matrix, y_supervised)
# print("total likelihood: {}".format(total_likelihood))
if self._stopping_time(old_likelihood, total_likelihood):
break
old_likelihood = total_likelihood.copy()
self.clf = clf
ys = pickle.load(open('binarized_ys.pkl', 'rb'))
print("Done.")
for i in range(0,2):
x_train, x_test, y_train, y_test = train_test_split(Xs[i], ys[i], test_size=1./3, random_state=3330)
features=len(x_train)
objects=len(y_train)
clf = BernoulliNB(alpha=0, binarize=0.0 , class_prior=None , fit_prior=True )
clf = clf.fit(x_train, y_train)
a= clf._joint_log_likelihood(x_train)
print("joint log likelyhood train")
print(a)
res=[]
for i in range(0,objects):
j=0
res.append([a[i][j]/(a[i][j]+a[i][j+1]),a[i][j+1]/(a[i][j]+a[i][j+1])])
sum=0
for i in range(0,objects):
if y_train[1].__eq__(False):
sum += res[i][0]
else:
sum += res[i][1];
#print("Alpha LIST : " ,alphaVal )
# For 10 datasets
for e in range(0, 10):
X_train, X_test, y_train, y_test = train_test_split(
Xs[e], ys[e], test_size=1. / 3, random_state=6099) # A20396099
# For 15 alpha values
for alp in range(0, 15):
# BernoulliNB classifier
clf = BernoulliNB(alpha=alphaVal[alp],
binarize=0.0,
fit_prior=True,
class_prior=None)
# fitting model on train data
clf.fit(X_train, y_train)
# prediction for train data using jll
predict_train = clf._joint_log_likelihood(X_train)
# prediction for test data using jll
predict_test = clf._joint_log_likelihood(X_test)
log_train, log_test = 0, 0
# print("Train : ", log_train)
# print("Train : ", log_train)
# print("Predict Train jll: ", predict_train)
# print("Predict Test jll: ", predict_test)
# summing test predections
for test in range(len(predict_test)):
if y_test[test] == True:
log_test += predict_test[test][1]
else:
log_test += predict_test[test][0]
# summing train predections
for train in range(len(predict_train)):
test_jll = np.zeros((10, 15))
for i in range(0, 10):
idx = 0
# Split datasets
x_train, x_test, y_train, y_test = train_test_split(Xs[i],
ys[i],
test_size=1. / 3,
random_state=7000)
for j in alphas:
# 1. Create new Bernoulli Naive Bayes model using alpha value
mod = BernoulliNB(alpha=j)
# Fit the model to the training set
mod.fit(x_train, y_train)
# Compute the joint log likelihood for the training set, store it train_jll 2d array
total_res = mod._joint_log_likelihood(x_train)
y_train_binary = y_train * 1
entry_val = 0
# Sum-up by matching true labels
for k in range(0, len(y_train)):
entry_val += total_res[k][y_train_binary[k]]
# Store result
train_jll[i][idx] = entry_val
# 2. Compute the joint log likelihood for the testing set, store it test_jll 2d array
total_res = mod._joint_log_likelihood(x_test)
y_test_binary = y_test * 1
entry_val = 0
# Sum-up by matching true labels
for k in range(0, len(y_test)):
entry_val += total_res[k][y_test_binary[k]]
test_jll[i][idx] = entry_val
distribution = []
for i in range(-7, 8):
distribution.append(10**i)
for i in range(10):
X_train, X_test, y_train, y_test = train_test_split(Xs[i],
ys[i],
test_size=1. / 3,
random_state=4435)
for j in range(15):
classifier = BernoulliNB(alpha=distribution[j])
classifier.fit(X_train, y_train)
train_Y_score = classifier._joint_log_likelihood(X_train)
individual_joint_likelihood = 0.0
for k in range(0, len(y_train)):
if y_train[k] == True:
individual_joint_likelihood += train_Y_score[k][1]
else:
individual_joint_likelihood += train_Y_score[k][0]
train_joint_likelihood[i][j] = individual_joint_likelihood
test_Y_score = classifier._joint_log_likelihood(X_test)
individual_joint_likelihood = 0.0
for k in range(0, len(y_test)):
if y_test[k] == True:
individual_joint_likelihood += test_Y_score[k][1]
else:
from sklearn.naive_bayes import BernoulliNB
Xs = pickle.load(open('binarized_xs.pkl', 'rb'))
ys = pickle.load(open('binarized_ys.pkl', 'rb'))
train_jll = np.zeros((10, 15))
test_jll = np.zeros((10, 15))
for i_dataset in range(10):
X, y = Xs[i_dataset], ys[i_dataset]
X_train, X_test, y_train, y_test = train_test_split(X,
y,
test_size=1. / 3,
random_state=1527)
y_train_indices = [0 if i == False else 1 for i in y_train]
y_test_indices = [0 if i == False else 1 for i in y_test]
for i_alpha in range(-7, 8):
clf = BernoulliNB(alpha=10**i_alpha)
clf.fit(X_train, y_train)
sum_train_jll, sum_test_jll = 0, 0
for i in range(len(y_train)):
sum_train_jll += clf._joint_log_likelihood(X_train)[i][
y_train_indices[i]]
for i in range(len(y_test)):
sum_test_jll += clf._joint_log_likelihood(X_test)[i][
y_test_indices[i]]
train_jll[i_dataset][i_alpha + 7] = sum_train_jll
test_jll[i_dataset][i_alpha + 7] = sum_test_jll
pickle.dump((train_jll, test_jll), open('result.pkl', 'wb'))
train_jll = np.zeros((10, 15))
test_jll = np.zeros((10, 15))
# Anumber A20406657
for i in range(len(Xs)):
X_train, X_test, y_train, y_test = train_test_split(
Xs[i], ys[i], test_size=1. / 3, random_state=int("6657"))
#print(X_train)
for j in range(len(alpha_list)):
sum_1 = 0
sum_2 = 0
clf = BernoulliNB(alpha=alpha_list[j],
binarize=0.0,
class_prior=None,
fit_prior=True)
clf.fit(X_train, y_train)
joint_X_train = clf._joint_log_likelihood(X_train)
joint_X_test = clf._joint_log_likelihood(X_test)
for k in range(0, len(joint_X_train)):
if y_train[k] == True:
sum_1 += joint_X_train[k][1]
else:
sum_1 += joint_X_train[k][0]
#print(y_train[k])
for m in range(0, len(joint_X_test)):
if y_test[m] == True:
sum_2 += joint_X_test[m][1]
else:
sum_2 += joint_X_test[m][0]
train_jll[i][j] = sum_1
test_jll[i][j] = sum_2
