# Verify the corpus (category)
if sys.argv[1] == "ham":
corpus = "1"
elif sys.argv[1] == "spam":
corpus = "2"
else:
print("Usage: " + sys.argv[0] + " {spam|ham} output_filename")
quit(1)
output_filename = sys.argv[2]
else:
print("Usage: " + sys.argv[0] + " {spam|ham} output_filename")
quit(1)
# Initialize the feature extractor
extractor = EmailFeatureExtractor(vector_file)
# Iterate through each line in stdin
for line in sys.stdin:
message_filename = line.strip()
# See if the file exists
if not os.path.exists(message_filename):
print("Error: \"" + message_filename + "\" does not exist")
continue
# Open the file
fp = open(message_filename, "rb")
# Parse the file as an email message
try:
def handle(self):
# Get the request line (first command)
request_line = str(self.rfile.readline().strip(), 'iso-8859-1')
# Separate the words
words = request_line.split()
self.command = words[0]
protocol = words[1]
# Make sure the command is valid
if self.command != "CHECK" and \
self.command != "SYMBOLS" and \
self.command != "REPORT" and \
self.command != "REPORT_IFSPAM" and \
self.command != "SKIP" and \
self.command != "PING" and \
self.command != "PROCESS" and \
self.command != "TELL":
# Send a bad response
self.sendResponse("USAGE", True, 100, "")
return
# See if we are dealing with a PING or a SKIP
if self.command == "PING" or self.command == "SKIP":
# Send a short OK response
self.wfile.write(b'SPAMD/1.1 0 EX_OK\r\n\r\n')
return
# Get the headers; (we only care about Content-length)
self.length = 0
while True:
raw_header = str(self.rfile.readline().strip(), 'iso-8859-1')
header = raw_header.split(':', 2)
if header[0].strip().lower() == 'content-length':
self.length = int(header[1].strip())
if not raw_header:
break
# Read the specified number of bytes into a temporary file
temp_file = tempfile.NamedTemporaryFile(dir=self.tmp_dir)
while self.length > 0:
# See how many bytes we are going to read
block_size = 4096
if self.length > block_size:
bytes_to_read = block_size
else:
bytes_to_read = self.length
# Read the block and write it to the temp file
temp_file.write(self.rfile.read(bytes_to_read))
self.length -= block_size
# Flush the file to disk
temp_file.flush()
# Extract the features from the email message
fp = open(temp_file.name, "rb")
# Parse the email message
try:
message = email.message_from_binary_file(fp)
except:
fp.close()
self.sendResponse("OK", True, 100,
"Message could not be parsed as a valid email")
return
# Extract information from the message for logging purposes
recepient = message["X-Envelope-To"]
subject = message["Subject"]
# Extract the features
extractor = EmailFeatureExtractor(self.vectors)
vector_count = extractor.vectorCount()
#vector_count = 236655
features = extractor.extract(message)
# Close the file, both the binary version and temp file version
fp.close()
temp_file.close()
# Write out the features to another temporary file
features_file = \
tempfile.NamedTemporaryFile(dir=self.tmp_dir)
features_file.write(b'1') # Corpus identifier (ham for now)
for feature_number in sorted(features.keys()):
if feature_number <= vector_count:
features_file.write(bytearray(" " + str(feature_number) + ":" +
str(features[feature_number]), 'iso-8859-1'))
features_file.write(b'\n')
features_file.flush()
# Call the classifier on the features
try:
command = [self.mallet, 'classify-svmlight', '--input',
features_file.name, '--classifier', self.filter_model,
'--output', '-']
output = subprocess.check_output(command, stderr=subprocess.STDOUT)
except:
self.sendResponse("TEMPFAIL", False, 0, "")
features_file.close()
return
# Close the features temporary file; this will delete the file
features_file.close()
# Interpret the probabilities
output_words = output.strip().split()
ham_probability = float(output_words[2]) * 100
spam_probability = float(output_words[4]) * 100
# Determine if the message is spam
if ham_probability >= spam_probability:
is_spam = False
message = "The content filter has determined your email is not spam"
else:
is_spam = True
message = "The content filter has determined your email is spam"
# Send the results
self.sendResponse("OK", is_spam, spam_probability, message)
#print(recepient, "|", subject, "|", ham_probability, "|", spam_probability)
# Log the entry
with open("/var/log/spam_filter/filter.log", "a") as log_file:
log_file.write(str(ham_probability) + "|" + str(spam_probability) + "|" + str(recepient) + "|" + (subject) + "\n")
if sys.argv[1] == "ham":
corpus = "1"
elif sys.argv[1] == "spam":
corpus = "2"
else:
print("Usage: " + sys.argv[0] + " {spam|ham} output_filename")
quit(1)
output_filename = sys.argv[2]
else:
print("Usage: " + sys.argv[0] + " {spam|ham} output_filename")
quit(1);
# Initialize the feature extractor
extractor = EmailFeatureExtractor(vector_file)
# Iterate through each line in stdin
for line in sys.stdin:
message_filename = line.strip()
# See if the file exists
if not os.path.exists(message_filename):
print("Error: \"" + message_filename + "\" does not exist")
continue
# Open the file
fp = open(message_filename, "rb")
# Parse the file as an email message
try: