def kFoldCrossValidation(all_training_data,
trainingphrases_to_sentiment,
vocabulary,
k=5,
classifier="linear"):
random.shuffle(all_training_data)
number_of_examples = len(all_training_data)
size_of_test_set = number_of_examples / k
accuracy = 0.0
for i in xrange(0, number_of_examples - size_of_test_set + 1,
size_of_test_set):
trainingdata = all_training_data[:i] + all_training_data[
i + size_of_test_set:]
testdata = all_training_data[i:i + size_of_test_set]
print "Building Feature Vectors..."
X_train, Y_train = getXandY(trainingdata,
vocabulary,
trainingphrases_to_sentiment,
tfidf=False,
L1normalization=False,
L2normalization=True)
X_test, Y_test = getXandY(testdata,
vocabulary,
trainingphrases_to_sentiment,
tfidf=False,
L1normalization=False,
L2normalization=True)
print "Training..."
if classifier == "linear":
model = svm.LinearSVC()
model.fit(X_train, Y_train)
elif classifier == "ovo":
model = svm.SVC()
model.fit(X_train, Y_train)
elif classifier == "NB":
model = MultinomialNB()
model.fit(X_train, Y_train)
elif classifier == "SGD":
model = linear_model.SGDClassifier()
model.fit(X_train, Y_train)
elif classifier == "centroid":
model = NearestCentroid()
model.fit(X_train, Y_train)
elif classifier == "perceptron":
model = Perceptron()
model.fit(X_train, Y_train)
elif classifier == "ridge":
model = RidgeClassifier()
model.fit(X_train, Y_train)
elif classifier == "levels":
#coarse classifier
print "Training top level"
Y_train_coarse = []
for y in Y_train:
if int(y) < 2:
Y_train_coarse.append("low")
elif int(y) == 2:
Y_train_coarse.append("2")
else:
Y_train_coarse.append("high")
model = LevelClassifier()
top = svm.LinearSVC()
top.fit(X_train, Y_train_coarse)
model.add_classifier(top, 1, "")
print "Training Low"
X_low, Y_low = getXandY(trainingdata,
vocabulary,
trainingphrases_to_sentiment,
tfidf=False,
L1normalization=False,
L2normalization=True,
separate="low")
low = svm.LinearSVC()
low.fit(X_low, Y_low)
model.add_classifier(low, 2, "low")
print "Training High"
X_high, Y_high = getXandY(trainingdata,
vocabulary,
trainingphrases_to_sentiment,
tfidf=False,
L1normalization=False,
L2normalization=True,
separate="high")
high = svm.LinearSVC()
high.fit(X_high, Y_high)
model.add_classifier(high, 2, "high")
print "Testing..."
accuracy += model.score(X_test, Y_test)
return accuracy / k
