def test1():
tokenizer = RegexpTokenizer(r'\w+')
spams = []
hams = []
load_data(hams, spams, 'res/test1.txt')
spam_words = []
ham_words = []
for spam in spams:
spam_words += tokenizer.tokenize(spam)
for ham in hams:
ham_words += tokenizer.tokenize(ham)
naive_bayes = NaiveBayes()
naive_bayes.load(ham_words, spam_words)
test_spams = []
test_hams = []
load_data(test_hams, test_spams, 'res/test1_check.txt')
spam_correct = 0
spam_incorrect = 0
for word in test_spams:
result = naive_bayes.is_positive(tokenizer.tokenize(word))
if result:
spam_incorrect += 1
else:
spam_correct += 1
print('spam:', 'correct', spam_correct, 'incorrect', spam_incorrect)
print('spam:', (spam_correct / (spam_incorrect + spam_correct)) * 100, '%')
ham_correct = 0
ham_incorrect = 0
for word in test_hams:
result = naive_bayes.is_positive(tokenizer.tokenize(word))
if result:
ham_correct += 1
else:
ham_incorrect += 1
print('ham:', 'correct', ham_correct, 'incorrect', ham_incorrect)
print('ham:', (ham_correct / (ham_incorrect + ham_correct)) * 100, '%')
def test2(is_from_begginning=True, training_percent=70):
tokenizer = RegexpTokenizer(r'\w+')
data = get_data('res/SMSSpamCollection.txt')
training_data_length = int((len(data) * training_percent) / 100)
if is_from_begginning:
training_data = data[:training_data_length]
test_data_length = len(data) - training_data_length
test_data = data[-test_data_length:]
else:
training_data = data[-training_data_length:]
test_data_length = len(data) - training_data_length
test_data = data[test_data_length:]
training_hams = []
training_spams = []
divide_data(training_data, training_hams, training_spams)
training_spam_words = []
training_ham_words = []
for ham in training_hams:
training_ham_words += tokenizer.tokenize(ham)
for spam in training_spams:
training_spam_words += tokenizer.tokenize(spam)
naive_bayes = NaiveBayes()
naive_bayes.load(training_ham_words, training_spam_words)
test_hams = []
test_spams = []
divide_data(test_data, test_hams, test_spams)
spam_correct = 0
spam_incorrect = 0
for word in test_spams:
result = naive_bayes.is_positive(tokenizer.tokenize(word))
if result:
spam_incorrect += 1
else:
spam_correct += 1
print('spam:', 'correct', spam_correct, 'incorrect', spam_incorrect)
print('spam:', (spam_correct / (spam_incorrect + spam_correct)) * 100, '%')
ham_correct = 0
ham_incorrect = 0
for word in test_hams:
result = naive_bayes.is_positive(tokenizer.tokenize(word))
if result:
ham_correct += 1
else:
ham_incorrect += 1
print('ham:', 'correct', ham_correct, 'incorrect', ham_incorrect)
print('ham:', (ham_correct / (ham_incorrect + ham_correct)) * 100, '%')
def __init__(self, directory = 'data/corpus2', spam = 'spam', ham = 'ham', limit = 1000): self.spam_path = os.path.join(os.getcwd(), directory, spam) self.ham_path = os.path.join(os.getcwd(), directory, ham) self.limit = limit self.classifier = NaiveBayes()
class Trainer(object):
# Initializes the object
# @param self Trainer object
# @param directory The directory that contains the training folders
# @param spam The sub directory for the spam class
# @param ham The sub directory for the ham class.
# @param limit The number of emails to be scanned
def __init__(self, directory = 'data/corpus2', spam = 'spam', ham = 'ham', limit = 1000):
self.spam_path = os.path.join(os.getcwd(), directory, spam)
self.ham_path = os.path.join(os.getcwd(), directory, ham)
self.limit = limit
self.classifier = NaiveBayes()
# A wrapper for the train_classifier function.
# @param self The trainer object
# @param verbose Depending on verbosity information will be printed
# @return The classifier object
def train(self, verbose = False):
self.train_classifier(self.spam_path,'spam', verbose)
self.train_classifier(self.ham_path,'ham', verbose)
return self.classifier
# Converts a document into tokens and extracts features as mentioned in README.md
# @param self The Trainer object
# @param text The text to be scanned
def extract_features(self, text):
features = []
tokens = text.split()
porter = stem.porter.PorterStemmer()
tokens = [token for token in tokens if token not in stopwords.words('english')]
link = re.compile('http[s]?://(?:[a-zA-Z]|[0-9]|[$-_@.&+]|[!*\(\),]|(?:%[0-9a-fA-F][0-9a-fA-F]))+')
for token in tokens:
if len(token.translate(None,string.punctuation)) < 3:
continue
if "." + token in mimetypes.types_map.keys():
features.append('ATTACHMENT')
elif token.isdigit():
features.append('NUMBER')
elif token.upper() == token:
features.append('ALL_CAPS')
features.append(porter.stem(token.translate(None, string.punctuation)).lower())
elif link.match(token):
features.append('LINK')
else:
features.append(porter.stem(token.translate(None, string.punctuation)).lower())
return features
# The function that does the actual classfication
# @param path The path of the data to be trained
# @param label The label underwhich the data is classified
# @param verbose the verbsoity of statistics printed
def train_classifier(self, path, label, verbose):
limit = len(os.listdir(path)) < self.limit and len(os.listdir(path)) or self.limit
if verbose:
print colored("Training %d emails in %s class" %(limit, label),'green')
os.chdir(path)
for email in os.listdir(path)[:self.limit]:
if verbose and verbose > 1:
print colored("Working on file %s" %(email),'green')
email_file = open(email, 'r')
email_text = email_file.read()
try:
email_text = bs4.UnicodeDammit.detwingle(email_text).decode('utf-8')
except:
print colored("Skipping file %s because of bad coding"%(email),'red')
continue
email_file.close()
email_text = email_text.encode('ascii', 'ignore')
features = self.extract_features(email_text)
self.classifier.train(features, label)
print colored(self.classifier,'green')
from classifier import NaiveBayes
from util import load_data
tokenizer = RegexpTokenizer(r'\w+')
spams = []
hams = []
load_data(hams, spams, 'res/SMSSpamCollection.txt')
spam_words = []
ham_words = []
for spam in spams:
spam_words += tokenizer.tokenize(spam)
for ham in hams:
ham_words += tokenizer.tokenize(ham)
naive_bayes = NaiveBayes()
naive_bayes.load(ham_words, spam_words)
message = ""
while message != "stop":
message = input("Enter your SMS:")
if naive_bayes.is_positive(tokenizer.tokenize(message)):
print("ham")
else:
print("spam")
folds = np.array_split(data, 10) #make 10 folds in the dataset
test_set = 0 #define test_set
#for each folds treat one fold as test set and 9 fols at train set
for y in range(len(folds)):
X_train = pd.DataFrame()
#if not test-set append fold in the train set
for x in range(len(folds)):
if x == test_set:
y_test = folds[x]['class'].values
X_test = folds[x].drop(['class'], axis=1)
else:
X_train = X_train.append(folds[x])
y_train = X_train['class'].values
X_train = X_train.drop(['class'], axis=1)
nb = NaiveBayes() #initialize Naive Bayes Classifier
nb.fit(X_train, y_train) #train model with train data
y_pred = nb.predict(X_test) #test model with test set
#find error with respect to zero-one loss function
error = nb.zero_one_loss_function(y_test, y_pred)
printstr = "\nAccuracy of 0-1 loss for fold {0} ::: {1}".format(
y, (1 - error))
print_both(file, printstr)
accuracy_list.append((1 - error))
#get mean square error
acc, precision, recall = nb.confusion_matrix(y_test, y_pred)
printstr = "\nCF for fold {0} ::: acc:: {1} :: precision:: {2} :: recall :: {3}".format(
y, acc, precision, recall)
print_both(file, printstr)
CF_accuracy_list.append(acc)
CF_precision_list.append(precision)