def fullConcWeight(weight, numSent):
goodWords, badWords = utils.getUniqueGoodandBadWords();
# posReviews = utils.loadAllTextFiles('dataset/txt_sentoken/pos/')
# negReviews = utils.loadAllTextFiles('dataset/txt_sentoken/neg/')
posReviews = utils.loadAllTextFiles('dataset/ebert_reviews/4-0/')
posReviews += utils.loadAllTextFiles('dataset/ebert_reviews/3-5/')
negReviews = utils.loadAllTextFiles('dataset/ebert_reviews/0-0/')
negReviews += utils.loadAllTextFiles('dataset/ebert_reviews/0-5/')
negReviews += utils.loadAllTextFiles('dataset/ebert_reviews/1-0/')
negReviews += utils.loadAllTextFiles('dataset/ebert_reviews/1-5/')
numberOfReviews = 950
posReviews = posReviews[:numberOfReviews]
negReviews = negReviews[:numberOfReviews]
correct = 0
for review in posReviews:
if conclusionWeight(review, weight, goodWords, badWords, numSent):
correct += 1
for review in negReviews:
if not conclusionWeight(review, weight, goodWords, badWords, numSent):
correct += 1
accuracy = correct / (len(posReviews) + len(negReviews))
return accuracy
def pipeline_test(num):
#posReviews = utils.loadAllTextFiles('dataset/txt_sentoken/pos/')
#negReviews = utils.loadAllTextFiles('dataset/txt_sentoken/neg/')
posReviews = utils.loadAllTextFiles('dataset/ebert_reviews/4-0/')
posReviews += utils.loadAllTextFiles('dataset/ebert_reviews/3-5/')
negReviews = utils.loadAllTextFiles('dataset/ebert_reviews/0-0/')
negReviews += utils.loadAllTextFiles('dataset/ebert_reviews/0-5/')
negReviews += utils.loadAllTextFiles('dataset/ebert_reviews/1-0/')
negReviews += utils.loadAllTextFiles('dataset/ebert_reviews/1-5/')
posReviews = posReviews[:950]
negReviews = negReviews[:950]
import random
import numpy as np
#random.seed()
accuracy = 0.0
percenttrain = 0.8
posTrainCount = int(percenttrain*len(posReviews))
negTrainCount = int(percenttrain*len(negReviews))
count = 0
for test in range(num):
count += 1
random.shuffle(posReviews)
random.shuffle(negReviews)
train_tups = [(r, 1) for r in posReviews[:posTrainCount]] + [(r, 0) for r in negReviews[:negTrainCount]]
random.shuffle(train_tups)
train_data = [tup[0] for tup in train_tups]
Y_train = np.array([tup[1] for tup in train_tups])
test_tups = [(r, 1) for r in posReviews[posTrainCount:]] + [(r, 0) for r in negReviews[negTrainCount:]]
random.shuffle(test_tups)
test_data = [tup[0] for tup in test_tups]
actual = np.array([tup[1] for tup in test_tups])
#pipe = Pipeline([('vect', CountVectorizer(stop_words=nltkstopwords.words('english'))),('tfidf', TfidfTransformer()),('clf', MultinomialNB())])
#pipe = Pipeline([('vect', CountVectorizer(stop_words=nltkstopwords.words('english'))),('tfidf', TfidfTransformer()),('clf', BernoulliNB())])
pipe = Pipeline([('vect', CountVectorizer(stop_words=nltkstopwords.words('english'))),('tfidf', TfidfTransformer()),('clf', LogisticRegression())])
#pipe = Pipeline([('vect', CountVectorizer(stop_words=nltkstopwords.words('english'))),('tfidf', TfidfTransformer()),('clf', SVC())])
pipe = pipe.fit(train_data, Y_train)
prediction = pipe.predict(test_data)
correct = 0
for i in range(len(actual)):
if(actual[i] == prediction[i]):
correct+=1
acc = correct / len(actual)
accuracy += acc
print("{:.3f} {:.3f}".format(acc, accuracy / count))
print("final accuracy: " + "{:.4f}".format(accuracy / num))
def fullPosNegTest():
goodWords, badWords = utils.getUniqueGoodandBadWords();
# posReviews = utils.loadAllTextFiles('dataset/txt_sentoken/pos/')
# negReviews = utils.loadAllTextFiles('dataset/txt_sentoken/neg/')
posReviews = utils.loadAllTextFiles('dataset/ebert_reviews/4-0/')
posReviews += utils.loadAllTextFiles('dataset/ebert_reviews/3-5/')
negReviews = utils.loadAllTextFiles('dataset/ebert_reviews/0-0/')
negReviews += utils.loadAllTextFiles('dataset/ebert_reviews/0-5/')
negReviews += utils.loadAllTextFiles('dataset/ebert_reviews/1-0/')
negReviews += utils.loadAllTextFiles('dataset/ebert_reviews/1-5/')
numberOfReviews = 950
posReviews = posReviews[:numberOfReviews]
negReviews = negReviews[:numberOfReviews]
correct = 0
count = 0
s = ""
for review in posReviews:
count += 1
if posminusneg(review, goodWords, badWords):
correct += 1
s = "correct!"
else:
s = "wrong :("
#if count % 10 == 0:
print(s + " {:.2f}% ".format(correct / count * 100) + str(count))
print("halfway there!")
for review in negReviews:
count += 1
if not posminusneg(review, goodWords, badWords):
correct += 1
s = "correct!"
else:
s = "wrong :("
#if count % 10 == 0:
print(s + " {:.2f}% ".format(correct / count * 100) + str(count))
return correct / count
def getSuperGoodBadAvg(iterations, topNum):
posList = ['JJ','NN','RB']
inclusion = True
# posReviews = utils.loadAllTextFiles('dataset/txt_sentoken/pos/')
# negReviews = utils.loadAllTextFiles('dataset/txt_sentoken/neg/')
# posposList = utils.loadPosList('dataset/txt_sentoken/negposlist.txt', posList, inclusion)
# negposList = utils.loadPosList('dataset/txt_sentoken/posposlist.txt', posList, inclusion)
posReviews = utils.loadAllTextFiles('dataset/ebert_reviews/4-0/')
posReviews += utils.loadAllTextFiles('dataset/ebert_reviews/3-5/')
negReviews = utils.loadAllTextFiles('dataset/ebert_reviews/0-0/')
negReviews += utils.loadAllTextFiles('dataset/ebert_reviews/0-5/')
negReviews += utils.loadAllTextFiles('dataset/ebert_reviews/1-0/')
negReviews += utils.loadAllTextFiles('dataset/ebert_reviews/1-5/')
posposList = utils.loadPosList('dataset/ebert_reviews/pos4-0.txt', posList, inclusion)
posposList += utils.loadPosList('dataset/ebert_reviews/pos3-5.txt', posList, inclusion)
negposList = utils.loadPosList('dataset/ebert_reviews/pos0-0.txt', posList, inclusion)
negposList += utils.loadPosList('dataset/ebert_reviews/pos0-5.txt', posList, inclusion)
negposList += utils.loadPosList('dataset/ebert_reviews/pos1-0.txt', posList, inclusion)
negposList += utils.loadPosList('dataset/ebert_reviews/pos1-5.txt', posList, inclusion)
numberOfReviews = 950
posReviews = posReviews[:numberOfReviews]
negReviews = negReviews[:numberOfReviews]
posposList = posposList[:numberOfReviews]
negposList = negposList[:numberOfReviews]
posTuples = list(zip(posReviews, posposList))
negTuples = list(zip(negReviews, negposList))
dataSetGoodWords, dataSetBadWords = utils.getUniqueGoodandBadWords()
totalacc = 0.0
for i in range(iterations):
totalacc += getSuperGoodBad(topNum, posTuples, negTuples, dataSetGoodWords, dataSetBadWords)
print("accuracy : " + "{:.4f}".format(totalacc / (i+1)))
