def extract_ngram_features(message, n=N_GRAM):
words = nltk.word_tokenize(message)
words = [word.lower() for word in words if word.isalnum()]
ws = set()
for i in xrange(n):
ws = ws.union(set([word for word in nltk.ingrams(words, i+1)]))
return {word: True for word in ws}
def __init__(self, order, string):
self.order = order
if self.order > 1:
self.backoff = MarkovModel(order - 1, string)
self.cfd = ConditionalFreqDist()
self.charset = self.backoff.charset
for ngram in ingrams(string, order):
context, char = ngram[:-1], ngram[-1]
self.cfd[context][char] += 1
else:
self.backoff = None
self.n = 0
self.fd = FreqDist(string)
self.charset = set(self.fd.keys())
def ngrams(l, n):
l = _remove_stopwords_lowercase(l)
phrase_list = []
for n_grams in ingrams(l, n):
phrase_list.append(' '.join(n_grams))
return phrase_list
import re, os, glob
from sys import argv
from nltk import FreqDist, PorterStemmer, ingrams
assert len(argv) == 4, "usage: %s inputdir outputdir dev|test" % argv[0]
assert os.path.isdir(argv[1])
indir = argv[1]
wordposngram = "%s/" % argv[2]
assert argv[3] in ("dev", "test")
devortest=argv[3]
leaves = re.compile(r" ([^ )]+)\)")
pos = re.compile(r"\(([^ ]+) [^ )]+\)")
porter = PorterStemmer()
print "extracting ngrams"
for train in glob.glob("%s/*.*.train" % indir):
fold = int(train.split(".")[-2])
if fold > 3: continue
wordpostrigrams = FreqDist(ingrams((porter.stem(word)+"/"+tag
for t in open(train)
for word, tag in zip(leaves.findall(t), pos.findall(t))), 3))
for test in glob.glob("%s/*/*.%d.%s*" % (indir, fold, devortest)):
output = "%s_%s" % (train.split("/")[-1], test.split("/")[-1])
testtrigrams = FreqDist(ingrams((porter.stem(word)+"/"+tag
for t in open(test).readlines()
for word,tag in zip(leaves.findall(t), pos.findall(t))), 3))
open(wordposngram+output, "w").writelines("%s\t%d\n" % (" ".join(a), b)
for a, b in testtrigrams.iteritems() if wordpostrigrams[a])
print output
print "done"
def ngram_phrases(t, n):
tokens = wordpunct_tokenize(t)
tokens = _remove_stopwords_lowercase(tokens)
return set(' '.join(n_grams) for n_grams in ingrams(tokens, n))
def ngram_phrases(t, n):
tokens = wordpunct_tokenize(t)
tokens = clean_words(tokens)
return [' '.join(n_grams) for n_grams in ingrams(tokens, n)]
from sys import argv
from nltk import FreqDist, PorterStemmer, ingrams
assert len(argv) == 4, "usage: %s inputdir outputdir dev|test" % argv[0]
assert os.path.isdir(argv[1])
indir = argv[1]
wordposngram = "%s/" % argv[2]
assert argv[3] in ("dev", "test")
devortest = argv[3]
leaves = re.compile(r" ([^ )]+)\)")
pos = re.compile(r"\(([^ ]+) [^ )]+\)")
porter = PorterStemmer()
print "extracting ngrams"
for train in glob.glob("%s/*.*.train" % indir):
fold = int(train.split(".")[-2])
if fold > 3: continue
wordpostrigrams = FreqDist(
ingrams((porter.stem(word) + "/" + tag for t in open(train)
for word, tag in zip(leaves.findall(t), pos.findall(t))), 3))
for test in glob.glob("%s/*/*.%d.%s*" % (indir, fold, devortest)):
output = "%s_%s" % (train.split("/")[-1], test.split("/")[-1])
testtrigrams = FreqDist(
ingrams(
(porter.stem(word) + "/" + tag for t in open(test).readlines()
for word, tag in zip(leaves.findall(t), pos.findall(t))), 3))
open(wordposngram + output,
"w").writelines("%s\t%d\n" % (" ".join(a), b)
for a, b in testtrigrams.iteritems()
if wordpostrigrams[a])
print output
print "done"