def compute_quality_for_corpus(corpus_dir):
truth_dic = methods.read_classification_from_file(methods.add_slash(corpus_dir)+"!truth.txt")
pred_dic = methods.read_classification_from_file(methods.add_slash(corpus_dir)+"!prediction.txt")
bc1 = BinaryConfusionMatrix('SPAM', 'OK')
bc1.compute_from_dicts(truth_dic, pred_dic)
dict_score = bc1.as_dict()
fn=dict_score['fn']
tn=dict_score['tn']
fp=dict_score['fp']
tp=dict_score['tp']
return quality_score(tp, tn, fp, fn), tp, tn, fp, fn
def train(self,path_to_truth_dir):
corpus = Corpus(path_to_truth_dir)
#Read truth file
truth = methods.read_classification_from_file(methods.add_slash(path_to_truth_dir)+"!truth.txt")
#Make truth global
self.truth = truth
for fname, body in corpus.emails_as_string():
email_as_file = open(methods.add_slash(path_to_truth_dir) + fname,'r',encoding = 'utf-8')
#Read email with EMAIL parser
msg = email.message_from_file(email_as_file)
self.extract_senders_list(msg,fname)
self.check_subject(msg,fname)
#Generate dict's
methods.generate_file_from_dict(self.path_bl , self.black_list)
methods.generate_file_from_dict(self.path_wl ,self.white_list)
methods.generate_file_from_dict(self.path_ssl , self.spam_subject_list)
methods.generate_file_from_dict(self.path_hsl ,self.ham_subject_list)
def test(self, path_to_test_dir):
predictions = {} #Predictions dict {fname:prediction}
bs = Bayesian.Bayesian()
corpus = Corpus(path_to_test_dir)
#Read dict's (if test called before train)
black_list_dict = methods.read_dict_from_file(self.path_bl)
white_list_dict = methods.read_dict_from_file(self.path_wl)
spam_subject_dict = methods.read_dict_from_file(self.path_ssl)
ham_subject_dict = methods.read_dict_from_file(self.path_hsl)
for fname, body in corpus.emails_as_string():
#Open email with parser
email_as_file = open(methods.add_slash(path_to_test_dir) + fname,'r',encoding = 'utf-8')
msg = email.message_from_file(email_as_file)
#Check if sender in a black list
if (self.extract_email_adress_from_text(msg['From']) in black_list_dict):
predictions[fname] = 'SPAM'
elif(self.extract_email_adress_from_text(msg['From']) in white_list_dict):
#Check if sender in a white list
predictions[fname] = 'OK'
#Check if subject in a black list
elif(self.extract_email_adress_from_text(msg['From']) in spam_subject_dict):
prediction[fname] = 'SPAM'
#Check if subject in a white list
elif(self.extract_email_adress_from_text(msg['From']) in ham_subject_dict):
prediction[fname] = 'OK'
#Run Bayesian checker
else:
if (bs.bayesian_prediction(methods.get_text(msg))) > 0.485:
predictions[fname] = 'SPAM'
else:
predictions[fname] = 'OK'
#Generate prediction file
bf = BaseFilter(path_to_test_dir,predictions)
bf.generate_prediction_file()
def generate_prediction_file(self):
path_to_prediction_file = methods.add_slash(self.path_to_dir)+'!prediction.txt'
prediction_file = open(path_to_prediction_file, 'w+')
for i in self.pred_dict.keys():
prediction_file.write(i + " " + self.pred_dict[i] + "\n")
def hams_as_string():
for file_name in os.listdir(self.path_to_dir):
if not file_name.startswith("!"):
if (is_ham(file_name)):
with io.open(methods.add_slash(self.path_to_dir)+file_name,'r', encoding ='utf-8') as body:
yield[file_name,body.read()]
def emails_as_string(self):
for file_name in os.listdir(self.path_to_dir):
if not file_name.startswith("!"):
with io.open(methods.add_slash(self.path_to_dir)+file_name,'r', encoding ='utf-8', errors='ignore') as body:
yield[file_name,body.read()]