def main(parser_dir,
text_loc,
out_dir,
codec="utf8",
profile=False,
debug=False):
if debug:
redshift.parser.set_debug(debug)
if not os.path.exists(out_dir):
os.mkdir(out_dir)
print "Loading parser"
parser = redshift.parser.Parser(parser_dir)
for i in range(1000):
with codecs.open(text_loc, 'r', 'utf8') as file_:
input_text = file_.read()
sentences = [
Input.from_pos(p.strip().encode(codec))
for i, p in enumerate(input_text.split('\n')) if p.strip()
]
t1 = time.time()
parse(parser, sentences)
t2 = time.time()
print '%d sents took %0.3f ms. %s mem' % (len(sentences),
(t2 - t1) * 1000.0,
mem())
def main(parser_dir, text_loc, out_dir, codec="utf8", profile=False, debug=False):
if debug:
redshift.parser.set_debug(debug)
if not os.path.exists(out_dir):
os.mkdir(out_dir)
print "Loading parser"
parser = redshift.parser.Parser(parser_dir)
with codecs.open(text_loc, 'r', 'utf8') as file_:
input_text = file_.read()
sentences = [Input.from_pos(p.strip().encode(codec)) for i, p in
enumerate(input_text.split('\n'))
if p.strip()]
if profile:
cProfile.runctx("parse(parser, sentences)",
globals(), locals(), "Profile.prof")
s = pstats.Stats("Profile.prof")
s.strip_dirs().sort_stats("time").print_stats()
else:
t1 = time.time()
parse(parser, sentences)
t2 = time.time()
print '%d sents took %0.3f ms' % (len(sentences), (t2-t1)*1000.0)
with open(os.path.join(out_dir, 'parses'), 'w') as out_file:
for sentence in sentences:
out_file.write(sentence.to_conll())
out_file.write('\n\n')
def main(model_dir,
train_loc,
dev_loc,
iters=5,
n_sents=0,
feat_thresh=5,
beam_width=4):
sent_strs = open(train_loc).read().strip().replace('|', '/').split('\n')
# Apply limit
if n_sents != 0:
sent_strs = sent_strs[:n_sents]
tagger = redshift.tagger.train('\n'.join(sent_strs),
model_dir,
beam_width=beam_width,
nr_iter=iters,
feat_thresh=feat_thresh)
dev_input = [
Input.from_pos(s.replace('|', '/'))
for s in open(dev_loc).read().strip().split('\n')
]
t = 1e-100
c = 0
for sent in dev_input:
gold_tags = [tok.tag for tok in sent.tokens]
tagger.tag(sent)
for i, token in enumerate(sent.tokens):
c += gold_tags[i] == token.tag
t += 1
print c / t
def main(parser_dir,
text_loc,
out_dir,
codec="utf8",
profile=False,
debug=False):
if debug:
redshift.parser.set_debug(debug)
if not os.path.exists(out_dir):
os.mkdir(out_dir)
print "Loading parser"
parser = redshift.parser.Parser(parser_dir)
with codecs.open(text_loc, 'r', 'utf8') as file_:
input_text = file_.read()
sentences = [
Input.from_pos(p.strip().encode(codec))
for i, p in enumerate(input_text.split('\n')) if p.strip()
]
if profile:
cProfile.runctx("parse(parser, sentences)", globals(), locals(),
"Profile.prof")
s = pstats.Stats("Profile.prof")
s.strip_dirs().sort_stats("time").print_stats()
else:
t1 = time.time()
parse(parser, sentences)
t2 = time.time()
print '%d sents took %0.3f ms' % (len(sentences), (t2 - t1) * 1000.0)
with open(os.path.join(out_dir, 'parses'), 'w') as out_file:
for sentence in sentences:
out_file.write(sentence.to_conll())
out_file.write('\n\n')
def main(parser_dir, text_loc, out_dir, codec="utf8", profile=False, debug=False):
if debug:
redshift.parser.set_debug(debug)
if not os.path.exists(out_dir):
os.mkdir(out_dir)
print "Loading parser"
parser = redshift.parser.Parser(parser_dir)
for i in range(1000):
with codecs.open(text_loc, 'r', 'utf8') as file_:
input_text = file_.read()
sentences = [Input.from_pos(p.strip().encode(codec)) for i, p in
enumerate(input_text.split('\n'))
if p.strip()]
t1 = time.time()
parse(parser, sentences)
t2 = time.time()
print '%d sents took %0.3f ms. %s mem' % (len(sentences), (t2-t1)*1000.0,
mem())
def postag(rvdata):
piece = rvdata[0][3]
test = TextBlob(piece)
print test
# print [str(x[0] + '/' + x[1]) for x in test.sentences[0].pos_tags]
tagstc = ' '.join(
[str(x[0]) + '/' + str(x[1]) for x in test.sentences[-1].pos_tags])
parser = redshift.parser.Parser('models')
# print tagstc
print test.sentences[0] + ''
testsent = Input.from_pos(
tagstc) #['A', 'list', 'of', 'tokens', 'is', 'required', '.'])
# testsent = Input.from_untagged(test.sentences[0].words)
parser.parse(testsent)
print testsent.to_conll() #.split('\n')
# sentence = Input.from_untagged(
# parser.parse(test)
'''
def main(model_dir, train_loc, dev_loc, iters=5, n_sents=0, feat_thresh=5, beam_width=4):
sent_strs = open(train_loc).read().strip().replace('|', '/').split('\n')
# Apply limit
if n_sents != 0:
sent_strs = sent_strs[:n_sents]
tagger = redshift.tagger.train('\n'.join(sent_strs), model_dir,
beam_width=beam_width, nr_iter=iters, feat_thresh=feat_thresh)
dev_input = [Input.from_pos(s.replace('|', '/'))
for s in open(dev_loc).read().strip().split('\n')]
t = 1e-100
c = 0
for sent in dev_input:
gold_tags = [tok.tag for tok in sent.tokens]
tagger.tag(sent)
for i, token in enumerate(sent.tokens):
c += gold_tags[i] == token.tag
t += 1
print c / t
def sentence():
from redshift.sentence import Input
return Input.from_pos('This/?? is/?? a/?? test/?? ./.')
def sentence():
from redshift.sentence import Input
return Input.from_pos('This/?? is/?? a/?? test/?? ./.')
def chunk(self,
text,
postagged=False,
sent_tokenized=False,
output_tags=False,
split_words=False):
if sent_tokenized:
sentences = text
else:
sentences = nltk.sent_tokenize(text)
if not postagged:
sentences = [
' '.join('/'.join(word_pos)
for word_pos in postagger.tag(sent))
for sent in sentences
]
else:
# Sentences are postagged. It can be ['sent/NN 1/CD ./.'] format (no change required) or
# [('sent','NN'), ('1','CD'), ('.','.')] (change to the earlier format is required)
if len(sentences) > 0 and not (isinstance(sentences[0], str) or
isinstance(sentences[0], unicode)):
sentences = [
' '.join('/'.join(word_pos) for word_pos in sent)
for sent in sentences
]
# Convert into Redshift sentence object
sentences = [Input.from_pos(sent) for sent in sentences]
for sentence in sentences:
# This will store the depparse result in each sentence object
self.parser.parse(sentence)
result = []
for sentence in sentences:
chunks = []
if split_words:
noun_phrase = []
else:
noun_phrase = ''
noun_head_idx = None
#length = sentence.length
for token in reversed(list(sentence.tokens)):
idx = token.id
word = token.word
pos = token.tag
parent = token.head
#rel = token.label
word = reverse_map_paren(word)
if word in RIGHT_PAREN:
continue
if parent == noun_head_idx and word not in LEFT_PAREN:
if output_tags:
if split_words:
noun_phrase[0:0] = (str(word), str(pos))
else:
noun_phrase = str(word) + '/' + str(
pos) + ' ' + noun_phrase
else:
if split_words:
noun_phrase[0:0] = str(word)
else:
noun_phrase = str(word) + ' ' + noun_phrase
else:
if noun_phrase:
chunks[0:0] = [noun_phrase]
noun_phrase = None
noun_head_idx = None
if pos.startswith('NN'):
if output_tags:
if split_words:
noun_phrase = [(str(word), str(pos))]
else:
noun_phrase = str(word) + '/' + str(pos)
else:
if split_words:
noun_phrase = [str(word)]
else:
noun_phrase = word
noun_head_idx = idx
if noun_phrase:
chunks[0:0] = [noun_phrase]
if sent_tokenized:
# Input is a list of sentences, output the chunks grouped by sentences
result.append(chunks)
else:
# Input is a plain text, output the chunks as one long list
result.extend(chunks)
return result