def classifyLR(train, test):
classifier.fit(train, twenty_train.target)
predicted = classifier.predict(test)
predicted_probs = classifier.predict_proba(test)
hlp.getStats(twenty_test.target, predicted)
hlp.plot_roc(twenty_test.target, predicted_probs[:, 1],
'Logistic Regression')
def classifyWithSVC(valC):
clf = svm.SVC(C=valC, probability=True, kernel='linear', random_state=42)
svdListTrain = td.getsvdListTrain()
nmfListTrain = td.getnmfListTrain()
svdListTest = td.getsvdListTest()
nmfListTest = td.getnmfListTest()
for min_df in [2,5]:
print ".......... With min_df = ", min_df , "..........."
if min_df == 2:
svd_matrix_train=svdListTrain[0]
nmf_matrix_train=nmfListTrain[0]
svd_matrix_test=svdListTest[0]
nmf_matrix_test=nmfListTest[0]
else:
svd_matrix_train=svdListTrain[1]
nmf_matrix_train=nmfListTrain[1]
svd_matrix_test=svdListTest[1]
nmf_matrix_test=nmfListTest[1]
print "With SVD"
clf.fit(svd_matrix_train, twenty_train.target)
predicted = clf.predict(svd_matrix_test)
probabilities = clf.predict_proba(svd_matrix_test)
hlp.getStats(twenty_test.target, predicted)
hlp.plot_roc(twenty_test.target, probabilities[:,1], 'SVM')
print "With NMF"
clf.fit(nmf_matrix_train, twenty_train.target)
predicted = clf.predict(nmf_matrix_test)
probabilitiesnmf = clf.predict_proba(nmf_matrix_test)
hlp.getStats(twenty_test.target, predicted)
hlp.plot_roc(twenty_test.target, probabilitiesnmf[:,1], 'SVM')
nmfListTest = td.getnmfListTest()
classifier = MultinomialNB()
for min_df in [2, 5]:
print "WIth min_df = ", min_df
if min_df == 2:
nmf_matrix_train = nmfListTrain[0]
nmf_matrix_test = nmfListTest[0]
tfidf_matrix_train = tfidfListTrain[0]
tfidf_matrix_test = tfidfListTest[0]
else:
nmf_matrix_train = nmfListTrain[1]
nmf_matrix_test = nmfListTest[1]
tfidf_matrix_train = tfidfListTrain[1]
tfidf_matrix_test = tfidfListTest[1]
print ".......... With SVD ........."
classifier.fit(tfidf_matrix_train, twenty_train.target)
predicted = classifier.predict(tfidf_matrix_test)
probabilities = classifier.predict_proba(tfidf_matrix_test)
hlp.getStats(twenty_test.target, predicted)
hlp.plot_roc(twenty_test.target, probabilities[:, 1], 'MultinomialNB')
print ".......... With NMF .........."
classifier.fit(nmf_matrix_train, twenty_train.target)
predicted = classifier.predict(nmf_matrix_test)
probabilities = classifier.predict_proba(nmf_matrix_test)
hlp.getStats(twenty_test.target, predicted)
hlp.plot_roc(twenty_test.target, probabilities[:, 1], 'MultinomialNB')
def classifyLR_WithReg(train, test):
l1_hyperPlaneParameters = []
l2_hyperPlaneParameters = []
l1_testErrors = []
l2_testErrors = []
l1_accu = []
l2_accu = []
l1_p = []
l2_p = []
l1_r = []
l2_r = []
for coeff in coeff_list:
print "************** ", coeff, "****************"
l1_classifier = LogisticRegression(penalty='l1',
C=coeff,
solver='liblinear')
l2_classifier = LogisticRegression(penalty='l2',
C=coeff,
solver='liblinear')
l1_classifier.fit(train, twenty_train.target)
l2_classifier.fit(train, twenty_train.target)
l1_predictions = l1_classifier.predict(test)
l2_predictions = l2_classifier.predict(test)
l1_predicted_probs = l1_classifier.predict_proba(test)
l2_predicted_probs = l2_classifier.predict_proba(test)
hlp.plot_roc(twenty_test.target, l1_predicted_probs[:, 1],
'LR with L1 regularization')
hlp.plot_roc(twenty_test.target, l2_predicted_probs[:, 1],
'LR with L2 regularization')
l1_testErrors.append(
100 -
smet.accuracy_score(twenty_test.target, l1_predictions) * 100)
l2_testErrors.append(
100 -
smet.accuracy_score(twenty_test.target, l2_predictions) * 100)
l1_accu.append(
smet.accuracy_score(twenty_test.target, l1_predictions) * 100)
l2_accu.append(
smet.accuracy_score(twenty_test.target, l2_predictions) * 100)
l1_p.append(
smet.precision_score(
twenty_test.target, l1_predictions, average='macro') * 100)
l2_p.append(
smet.precision_score(
twenty_test.target, l2_predictions, average='macro') * 100)
l1_r.append(
smet.recall_score(
twenty_test.target, l1_predictions, average='macro') * 100)
l2_r.append(
smet.recall_score(
twenty_test.target, l2_predictions, average='macro') * 100)
l1_hyperPlaneParameters.append(np.mean(l1_classifier.coef_))
l2_hyperPlaneParameters.append(np.mean(l2_classifier.coef_))
plotGraph(l1_testErrors, "l1")
plotGraph(l2_testErrors, "l2")
index = 0
for coeff in coeff_list:
print "coeff=", coeff
print "Test error l1=", l1_testErrors[index]
print "Mean of coeff l1=", l1_hyperPlaneParameters[index]
print "Accuracy l1=", l1_accu[index]
print "Precision l1=", l1_p[index]
print "Recall l1=", l1_r[index]
print "coeff=", coeff
print "Test error l2=", l2_testErrors[index]
print "Mean of coeff l2=", l2_hyperPlaneParameters[index]
print "Accuracy l2=", l2_accu[index]
print "Precision l2=", l2_p[index]
print "Recall l2=", l2_r[index]
index += 1
for min_df in [2, 5]:
print "Calculating for min_df = ", min_df
if min_df == 2:
svd_matrix_train = svdListTrain[0]
svd_matrix_test = svdListTest[0]
nmf_matrix_train = nmfListTrain[0]
nmf_matrix_test = nmfListTest[0]
else:
svd_matrix_train = svdListTrain[1]
svd_matrix_test = svdListTest[1]
nmf_matrix_train = nmfListTrain[1]
nmf_matrix_test = nmfListTest[1]
print "************With SVD***********"
clf_5fold_svd = callClassifier(svd_matrix_train)
clf_5fold_svd.fit(svd_matrix_train, twenty_train.target)
predicted = clf_5fold_svd.predict(svd_matrix_test)
probabilities = clf_5fold_svd.predict_proba(svd_matrix_test)
hlp.getStats(twenty_test.target, predicted)
hlp.plot_roc(twenty_test.target, probabilities[:, 1],
'SVM with cross-validation')
print "************With NMF***********"
clf_5fold_nmf = callClassifier(nmf_matrix_train)
clf_5fold_nmf.fit(nmf_matrix_train, twenty_train.target)
predictednmf = clf_5fold_nmf.predict(nmf_matrix_test)
probabilitiesnmf = clf_5fold_nmf.predict_proba(nmf_matrix_test)
hlp.getStats(twenty_test.target, predictednmf)
hlp.plot_roc(twenty_test.target, probabilitiesnmf[:, 1],
'SVM with cross-validation')