#!/usr/bin/env python

# xmltraintest.py v0.02 (March 23, 2004)
# by Joseph Connors

import glob, sys, string, re, getopt, time
from xmlrpclib import ServerProxy

def help(message):
	if message == 0:
		msg = "Usage: python " + sys.argv[0] + " [options] <archive dir>\n"
		msg += "\n"
		msg += "Options:\n"
		msg += "  -a   Set training mode to Train Always (default is Train Only Errors)\n"
		msg += "  -c   Clear exising corpus if it exists\n"
		msg += "  -d   Display detailed information about buckets\n"
		msg += "  -r#  Display row accuracy every # rows (0 turns off row accuracy display)\n"
		msg += "  -t   Test archive against current corpus (no changes made to corpus)\n"
		msg += "  -?   Display this help message\n"
		msg += "\n"
		msg += "Warning: You should backup your corpus before using this program.\n"
		msg += "         It will modify or replace your existing corpus."
	if message == 1:
		msg = "Error: Unable to connect to POPFile.\n"
		msg += "  Make sure POPFile is running.\n"
		msg += "  You must also have the XMLRPC.pm module installed in POPFile\n"
		msg += "  and set xmlrpc_enabled to 1 on POPFile's Advanced page."
	return msg

def compare_mf(a, b):					# Python version of POPFile function to compare file names (sort order reversed).
	regex = re.compile("popfile(\d+)=(\d+)\.msg")
	ad,am = regex.findall(a)[0]
	bd,bm = regex.findall(b)[0]

	if (ad == bd):
		return cmp(int(am), int(bm))
	else:
		return cmp(int(ad), int(bd))

if sys.argv[1:]:
	try:							# get argument from command line
		opts, path = getopt.gnu_getopt(sys.argv[1:],"acdr:t?")

		TrainMethod = "toe"				# default to TOE
		ClearCorpus = 0					# default use existing corps if it exists
		Detail = 0					# default does not display bucket details
		Row = 2						# default display row accuracy over every 2 rows (100 messages)

		for opt in opts:
			if opt[0] == "-a":
				TrainMethod = "ta"
			if opt[0] == "-c":
				ClearCorpus = 1
			if opt[0] == "-d":
				Detail = 1
			if opt[0] == "-r":
				Row = int(opt[1])
			if opt[0] == "-t":			# Testing Only disables TA and clear corpus options
				TrainMethod = "test"
				ClearCorpus = 0
			if opt[0] == "-?":
				sys.exit(help(0))

		if len(path) > 1:
			raise getopt.GetoptError("Too many parameters: [%s]" % string.join(path), None)
		if len(path) == 0:
			raise getopt.GetoptError("Archive directory must be given", None)
		path = path[0]

	except getopt.error, detail:
		print "Error:", detail
		sys.exit(1)
else:
	sys.exit(help(0))

try:
	POPFile = ServerProxy("http://localhost:8081")	# connect to POPFile
	pf = POPFile.Classifier.Bayes			# just a naming shortcut
	key = pf.get_session_key("admin","")		# get session key, send username (admin in PF v0.21) and password
except:
	sys.exit(help(1))

try:
	try:
		errorfile = file("errors.txt","w")
	except:
		print "Error: can not write errors.txt, continuing anyway."

	maillist = []					# create list

	if ClearCorpus:					# clear existing corpus, retain buckets and magnets
		print "Clearing existing corpus"
		for each in pf.get_buckets(key):
			pf.clear_bucket(key,each)

	archive = glob.glob1(path,"*")			# get archive bucket folders
	if len(archive) == 0:
		sys.exit("Error: No bucket folders found in " + path)
	for folder in archive:				# create buckets from archive folders
		if not folder in pf.get_buckets(key):
			if not TrainMethod == "test":
				pf.create_bucket(key,folder)
			else:
				print "Warning: Archive bucket", folder, "was not found in corpus."
				print "         Testing will not create it, this may cause misclassification."
		for mail in glob.glob1(path + "/" + folder,"popfile*.msg"):
			maillist += [folder + "/" + mail]
	maillist.sort(compare_mf)			# sort messages by file name

	print len(maillist), "messages found in", len(archive), "buckets"

	errors = 0
	unclassified = 0
	correct = 0
	count = 0
	rowerrors = 0
	reclassified = 0

	cputime = time.clock()				# start cpu clock time
	for mail in maillist:
		bucket = mail.split("/")[0]
		result = pf.classify(key, path + "/" + mail)
		if result != bucket:
			if not TrainMethod == "test":
				pf.add_message_to_bucket(key, bucket, path + "/" + mail)
			if result == "unclassified":
				unclassified += 1	# count unclassified messages
				sys.stdout.write('?')
			else:
				errors += 1		# count misclassified messages
				rowerrors += 1		# count misclassified for row
				sys.stdout.write('-')
			reclassified += 1
			errorfile.write(mail + "\t" + result + "\n")
		else:
			correct += 1
			sys.stdout.write('+')
			if TrainMethod == "ta":
				pf.add_message_to_bucket(key, bucket, path + "/" + mail)
				reclassified += 1
		count += 1

		if Row:					# print row statistics
			if (count % (50*Row) == 0):
				print " %.0f" % ((float)((50*Row)-rowerrors)/(50*Row)*100) + "%",
				if TrainMethod == "test":
					print "missed",
				else:
					print "trained",
				print reclassified,
				print "(%.1f" % ((float)(correct+unclassified)/(correct+unclassified+errors)*100) + "%)"
				rowerrors = 0
				sys.stdout.flush()
				errorfile.flush()
			elif (count % 50 == 0):
				print
		elif (count % 79 == 0):
			print
			sys.stdout.flush()
			errorfile.flush()

except KeyboardInterrupt:
	print "\n*Exited Early by user*"

seconds = time.clock() - cputime

print
if TrainMethod == "toe":
	print "Train Only Errors\n"
elif TrainMethod == "ta":
	print "Train Always\n"
else:
	print "Testing Only (no training)\n"

print "Processed", correct+unclassified+errors, "messages in", round(seconds,2), "seconds."
print "Misclassified:\t", errors
print "Unclassified:\t", unclassified
print "Correct:\t", correct
print
if correct+unclassified+errors > 0:
	print "%.2f" % ((float)(correct+unclassified)/(correct+unclassified+errors)*100) + "% accuracy"

if Detail:
	print
	print "Bucket word counts:"

	space = 0
	for bucket in archive:
		x = len(bucket[0:40])
		if x > space:
			space = x
	fmtst = (" %" + space.__str__() + "s")
	
	print fmtst % (""),
	print "%10s %10s %10s" % ("Unique", "Total", "Messages")

	magnets = pf.get_buckets_with_magnets(key)

	for bucket in archive:
		print fmtst % (bucket[0:40]),
		print "%10.1d" % (pf.get_bucket_unique_count(key,bucket)),
		print "%10.1d" % (pf.get_bucket_word_count(key,bucket)),
		print "%10.1d" % (len(glob.glob1(path + "/" + bucket,"popfile*.msg"))),
		if bucket in magnets:
			print "  magnet",
		print
	print fmtst % (""),
	print "%10s" % ("--------"),
	print "%10s" % ("--------"),
	print "%10s" % ("--------")

	print fmtst % (""),
	print "%10d" % (pf.get_unique_word_count(key,bucket)),
	print "%10d" % (pf.get_word_count(key,bucket)),
	print "%10d" % (len(maillist))

pf.release_session_key(key)