def train_sklearn_classifier(self):
if not self.feature_words:
self._read_csv()
self._generate_word_features()
shuffle(self.documents)
feature_sets = [(self.__document_features(tok), lab)
for (tok, lab) in self.documents]
cutoff = int(len(feature_sets) * self.test_ratio)
train_set, test_set = feature_sets[cutoff:], feature_sets[:cutoff]
print("Totals({0}) Training({1}) Test({2})".format(
len(feature_sets), len(train_set), len(test_set)))
self.classifier = DecisionTreeClassifier.train(test_set)
print('Achieved {0:.2f}% accuracy against training set'.format(
nltk.classify.accuracy(self.classifier, train_set) * 100))
print('Achieved {0:.2f}% accuracy against test set'.format(
nltk.classify.accuracy(self.classifier, test_set) * 100))
vocabulary.add(word)
def format_dataset(fileids, featureSet):
dataset = list()
for fileid in fileids:
features = dict()
for word in featureSet:
features[word] = word in movie_reviews.words(fileid)
pos_or_neg = fileid[:3]
example = (features, pos_or_neg)
dataset.append(example)
return dataset
# Get the datasets ready
train_set = format_dataset(train_fileids, vocabulary)
test_set = format_dataset(test_fileids, vocabulary)
# Create some classifiers
from nltk.classify.decisiontree import DecisionTreeClassifier
tree = DecisionTreeClassifier.train(train_set)
from nltk.classify.naivebayes import NaiveBayesClassifier
bayes = NaiveBayesClassifier.train(train_set)
# Test the classifiers
from nltk.classify import accuracy
print("Decision Tree accuracy: ", accuracy(tree, test_set))
print("Naive Bayes accuracy: ", accuracy(bayes, test_set))
def f(x): return DecisionTreeClassifier.train(x, binary=True, verbose=True)
('bag_of_words',
Pipeline([
('selector', FeatureGetter(train_entries)),
('vect',
CountVectorizer(ngram_range=(1, 1),
max_features=500)),
('tfidf', TfidfTransformer(use_idf=True)),
])),
# Weights for the features
],
transformer_weights={
'feature_1': 1.5,
'feature_2': 0.5,
'feature_3': 0.5,
'feature_4': 1.5,
},
)),
# Decision tree used as classifier
('clf', DecisionTreeClassifier(max_features=500)),
])
pipeline_pickle_path = 'data/pipeline.pickle'
pickle.dump(pipeline, open(pipeline_pickle_path, 'wb'))
print("About to run the pipeline...")
train_data = sp.convertDataToList(sp.train)
test_data = sp.convertDataToList(sp.train)
train_entries, train_langs = sp.returnEntriesWithSpoken(train_data)
test_entries, test_langs = sp.returnEntriesWithSpoken(test_data)
pipeline.fit(train_entries, train_langs)
y = pipeline.predict(train_entries)
print(classification_report(y, test_langs))
if (len(elem[0]) > 3):
negative.append(elem)
combined = list(set(positive + negative))
feature_set = []
for i, elem in enumerate(total_text):
# elem is the words in a document
features = {}
for word, number in combined:
features['contains({})'.format(word)] = (word in elem)
pos_or_neg = "positive" if (stars_array[i] >= 3) else "negative"
feature_set.append((features, pos_or_neg))
print "Finished Featurization. It took " + str(time.time() - start_time) + " to run."
cutoff = int(len(feature_set) * 3/4)
training = feature_set[:cutoff]
testing = feature_set[cutoff:]
classifier = DecisionTreeClassifier.train(training, binary=True, verbose=True)
count = 0
for elem in testing:
val = classifier.classify(elem[0])
if val == elem[1]:
count += 1
print float(count)/len(testing)
