def best_n_attack(classifier):
WORD_LIMIT = 200
goodWords = run_best_n_words(classifier)
spam = load_spam()
avg = 0
flag = False
import random
for comment in spam:
for i in range(WORD_LIMIT):
comment += " {}".format(random.choice(goodWords))
if type(classifier) == NaiveClassifier:
if classifier.predict(comment) == False:
avg += i
break
if i == WORD_LIMIT - 1:
avg += WORD_LIMIT
else:
if classifier.classify(list_to_dict(
comment.split(" "))) == str(0):
avg += i
break
if i == WORD_LIMIT - 1:
avg += WORD_LIMIT
print(avg / len(spam))
def find_witness(spam, legit, classifier):
length = min([len(spam), len(legit)])
curr = spam
i = 0
flag = False
prev = curr
if type(classifier) == NaiveClassifier:
while classifier.predict("".join(curr)) != True:
prev = curr
for j in range(len(curr)):
if curr[j] not in spam:
del curr[j]
flag = True
break
if not flag:
for j in range(len(spam)):
if spam[j] not in curr:
curr.append(j)
break
flag = False
i += 1
if i == length:
return curr, prev
elif type(classifier) == nltk.classify.maxent.MaxentClassifier:
while classifier.classify(list_to_dict(curr)) != str(1):
prev = curr
for j in range(len(curr)):
if curr[j] not in spam:
del curr[j]
flag = True
break
if not flag:
for j in range(len(spam)):
if spam[j] not in curr:
curr.append(j)
break
flag = False
i += 1
if i == length:
return curr, prev
else:
while classifier.classify("".join(curr)) != 0:
prev = curr
for j in range(len(curr)):
if curr[j] not in spam:
del curr[j]
flag = True
break
if not flag:
for j in range(len(spam)):
if spam[j] not in curr:
curr.append(j)
break
flag = False
i += 1
if i == length:
return curr, prev
return (curr, prev)
def first_n_words(spam, legit, classifier, wordlist):
L = []
to_spam, _ = find_witness(spam, legit, classifier)
for word in wordlist:
to_spam.append(word)
if type(classifier) == NaiveClassifier:
if classifier.predict("".join(to_spam)) == False:
L.append(word)
to_spam = to_spam[:len(to_spam)-1]
else:
if classifier.classify(list_to_dict(spam)) == str(0):
L.append(word)
to_spam = to_spam[:len(to_spam)-1]
if len(L) == 20:
break
return L
def trick_maxent(classifier):
WORD_LIMIT = 200
goodWords = load_goodwords('Maxent_Data/goodwords.txt')
spam = load_spam()
spam = [line.split(" ")[1] for line in spam]
avg = 0
flag = False
import random
for comment in spam:
for i in range(0, WORD_LIMIT):
comment += " {}".format(random.choice(goodWords))
if classifier.classify(list_to_dict(comment)) == str(0):
avg += i
break
if i == WORD_LIMIT - 1:
avg += WORD_LIMIT
print(avg / len(spam))
def test(filename, classifier, type="maxent"):
testing_set = []
with open(filename, "r") as f:
reader = csv.reader(f)
for row in reader:
if row[0] == 'COMMENT_ID':
continue
content = re.sub(r'[^\w\s]', '', row[3]).lower()
content = content.split(" ")
testing_set.append([content, row[4]])
right = 0
wrong = 0
wrong_items = []
spam = []
for item in testing_set:
if type == "maxent":
if classifier.classify(list_to_dict(item[0])) == item[1]:
right += 1
else:
wrong += 1
wrong_items.append([item[0], item[1]])
else:
string = ""
for word in item[0]:
string += "{} ".format(word)
if classifier.predict(string) == int(item[1]):
if int(item[1]) == 1:
spam.append(string)
right += 1
else:
wrong += 1
wrong_items.append([item[0], item[1]])
print(f'right: {right}, wrong: {wrong}')
if type != "maxent":
with open("spam.txt", "w") as f:
for comment in spam:
f.write("{}\n".format(comment))