def main(args):
if os.path.exists(args.output_file):
raise FileExistsError('File exists: {}'.format(args.output_file))
csv_reader = pd.read_csv(args.input_file,
chunksize=args.batch_size,
usecols=[
'SUBJECT_ID', 'HADM_ID', 'CHARTDATE',
'CATEGORY', 'DESCRIPTION', 'ISERROR', 'TEXT'
],
dtype={
'SUBJECT_ID': int32,
'HADM_ID': 'str',
'CATEGORY': 'str',
'DESCRIPTION': 'str',
'ISERROR': 'str',
'TEXT': 'str'
},
keep_default_na=False,
na_values='')
with jsonlines.open(args.output_file, 'w') as notes_tokenized_file:
tokenizer = Tokenizer(args.batch_size, args.n_cpus, args.n_threads,
MODE)
for i, notes_batch in enumerate(csv_reader):
process_batch(notes_batch, i, tokenizer, notes_tokenized_file)
def load(cls, path, fields, tokenizer_lang, tokenizer_dir, use_gpu, verbose=True, max_sent_length=math.inf):
tokenizer = Tokenizer(lang=tokenizer_lang, dir=tokenizer_dir, use_gpu=use_gpu, verbose=True)
sentences = []
fields = [field if field is not None else Field(str(i))
for i, field in enumerate(fields)]
with open(path, 'r') as f:
lines = []
for line in tokenizer.format(tokenizer.predict(f.read())):
line = line.strip()
if not line:
if len(lines) > max_sent_length:
logger.info('Discarded sentence longer than max_sent_length:',
len(lines), file=sys.stderr)
lines = []
continue
sentences.append(Sentence(fields, lines))
lines = []
else:
if not line.startswith('#'):
# append fake columns
line = '{}\t{}'.format(line, '\t'.join(['_' for i in range(len(CoNLL._fields) - len(line.split('\t')))]))
assert len(CoNLL._fields) == len(line.split('\t')), '{} - {} vs {}'.format(line, len(CoNLL._fields), len(line.split()))
lines.append(line)
return cls(fields, sentences)
def __init__(self,
NLG_param_dir,
NLG_model_fname,
tokenizer,
NLU_param_dir=None,
NLU_model_fname=None):
self.tokenizer = Tokenizer(tokenizer, '../tokenizer/e2e.model')
self.tokenizer_mode = tokenizer
self.device = torch.device(
'cuda' if torch.cuda.is_available() else 'cpu')
saved_data = torch.load(
NLG_param_dir.rstrip('/') + '/' + NLG_model_fname)
self.model_NLG = saved_data['model']
f = open(NLG_param_dir.rstrip('/') + '/dictionary.json',
'r',
encoding='utf-8')
self.dictionary = json.load(f)
f.close()
# beam-search settings
self.n_beam = 5
# NLU
if (NLU_param_dir is not None) and (NLU_model_fname is not None):
self.NLU = NLU(NLU_param_dir, NLU_model_fname, tokenizer)
else:
self.NLU = None
def initialize_tokenizer(vocabulary_path, is_bpe=True):
"""Initialize vocabulary from file.
We assume the vocabulary is stored one-item-per-line, so a file:
dog
cat
will result in a vocabulary {"dog": 0, "cat": 1}, and this function will
also return the reversed-vocabulary ["dog", "cat"].
Args:
vocabulary_path: path to the file containing the vocabulary.
Returns:
a pair: the vocabulary (a dictionary mapping string to integers), and
the reversed vocabulary (a list, which reverses the vocabulary mapping).
Raises:
ValueError: if the provided vocabulary_path does not exist.
"""
if is_bpe:
return bpe_tokenizer
if os.path.isfile(vocabulary_path):
rev_vocab = []
with open(vocabulary_path, mode="r") as f:
rev_vocab.extend(f.readlines())
rev_vocab = [line.strip() for line in rev_vocab]
return Tokenizer(_UNK, vocab_list=rev_vocab)
else:
raise ValueError("Vocabulary file %s not found.", vocabulary_path)
def test_corpus_load(self):
tokenizer = Tokenizer(**self.args)
raw_text_file = '/project/piqasso/Collection/IWPT20/train-dev/UD_Italian-ISDT/it_isdt-ud-dev.txt'
with open(raw_text_file) as fin:
for line in tokenizer.format(tokenizer.predict(fin.read())):
if line and not line.startswith('#'):
assert len(line.split('\t')) == 2, line
def get_embedding_fn(self, max_length=12):
self.max_length = max_length
self.s = Simplifier('tokenizer/zh_mapping.txt')
self.t = Tokenizer('tokenizer/spiece_all_bpe/spiece.all.bpe.130000.lower.model',
'tokenizer/lg.all.voc',
max_length
)
self.max_length = max_length
return self.embedding
def test_corpus_load(self):
tokenizer = Tokenizer(**self.args)
sin = io.StringIO(
"Un corazziere contro Scalfaro. L'attore le disse baciami o torno a riprendermelo."
)
for line in tokenizer.format(tokenizer.predict(sin.read())):
if line and not line.startswith('#'):
assert len(line.split('\t')) == 10, line
def predict(self,
data,
pred=None,
buckets=8,
batch_size=5000,
prob=False,
**kwargs):
r"""
Parses the data and produces a parse tree for each sentence.
Args:
data (str or list[list[str]]): input to be parsed: either
- a str, that will be tokenized first with the tokenizer for the parser language
- a path to a file to be read, either in CoNLL-U format or in plain text if :param text: is supplied.
- a list of lists of tokens
text (str): optional, specifies that the input data is in plain text in the specified language code.
pred (str or file): a path to a file where to write the parsed input in CoNLL-U fprmat.
bucket (int): the number of buckets used to group sentences to parallelize matrix computations.
batch_size (int): group sentences in batches.
prob (bool): whther to return also probabilities for each arc.
Return:
a Dataset containing the parsed sentence trees.
"""
args = self.args.update(locals())
init_logger(logger, verbose=args.verbose)
self.transform.eval()
if args.prob:
self.transform.append(Field('probs'))
if isinstance(data, str) and (not conll_format(data) or args.text):
self.transform.reader = Tokenizer(args.text,
dir=args.cache_dir,
verbose=args.verbose).reader()
logger.info("Loading the data")
dataset = Dataset(self.transform, data)
dataset.build(args.batch_size, args.buckets)
logger.info(f"\n{dataset}")
logger.info("Making predictions on the dataset")
start = datetime.now()
preds = self._predict(dataset.loader)
elapsed = datetime.now() - start
for name, value in preds.items():
setattr(dataset, name, value)
if pred is not None and is_master():
logger.info(f"Saving predicted results to {pred}")
self.transform.save(pred, dataset.sentences)
logger.info(
f"{elapsed}s elapsed, {len(dataset) / elapsed.total_seconds():.2f} Sents/s"
)
return dataset
def __init__(self, param_dir, model_fname, tokenizer):
self.tokenizer = Tokenizer(tokenizer, '../tokenizer/e2e.model')
self.tokenizer_mode = tokenizer
self.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
saved_data = torch.load(param_dir.rstrip('/')+'/'+model_fname)
self.model = saved_data['model']
f = open(param_dir.rstrip('/')+'/dictionary.json', 'r', encoding='utf-8')
self.dictionary = json.load(f)
f.close()
# beam-search settings
self.n_beam = 5
def test_download_resources(self):
self.assertTrue(not os.path.exists(self.MODEL_DIR))
tokenizer = Tokenizer(**self.args)
self.assertTrue(
os.path.exists(self.args['dir'])
and not os.path.isfile(self.args['dir']))
self.assertTrue(
os.path.exists(os.path.join(self.args['dir'], self.args['lang'])))
self.assertTrue(
os.path.exists(
os.path.join(self.args['dir'], self.args['lang'], 'tokenize')))
def test_download_resources(self):
tokenizer = Tokenizer(self.args['lang'])
self.assertTrue(os.path.isdir(self.MODEL_DIR))
self.assertTrue(
os.path.exists(
os.path.join(self.MODEL_DIR, 'tokenizer', self.args['lang'])))
self.assertTrue(
os.path.exists(
os.path.join(self.MODEL_DIR, 'tokenizer', self.args['lang'],
'tokenize')))
def main():
dataset = data_loader.load_text_file("data_2.txt")
tokenizer = Tokenizer()
separated = tokenizer.tokenize([dataset])
morfeusz = MorfeuszWrapperLexeme()
for sentence in separated:
analysed = morfeusz.analyse([w for w, tag in sentence])
print(analysed)
for word, analysis in analysed.items():
print("{}:".format(word))
print_analysis(analysis)
print()
def tokenize(path):
t = Tokenizer(path)
tokens = t.raw_tokenize()
idents = []
for token in tokens:
if token.kind.name == "IDENTIFIER":
name = token.spelling.lower()
name = re.sub("_", "", name)
idents.append(name)
return "\n".join(idents)
def main():
text = 'Charakteryzował się on ustawieniem zawodników w kształcie piramidy' \
' – bramkarz - 2 obrońców - 3 pomocników - 5 napastników (1-2-3-5).'
morfeusz = MorfeuszWrapperLexeme()
tokenizer = Tokenizer()
text = tokenizer.tokenize([text])
for sen in text:
analysed = morfeusz.analyse([w for w, tag in sen], as_xml=False)
print(analysed)
analysed = morfeusz.analyse([w for w, tag in sen], as_xml=True)
print(analysed)
print()
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
if self.args.feat in ('char', 'bert'):
self.WORD, self.FEAT = self.transform.FORM
else:
self.WORD, self.FEAT = self.transform.FORM, self.transform.CPOS
self.ARC, self.REL = self.transform.HEAD, self.transform.DEPREL
self.puncts = torch.tensor([
i for s, i in self.WORD.vocab.stoi.items() if ispunct(s)
]).to(self.args.device)
if getattr(self.args, 'text', None):
self.transform.reader = Tokenizer(self.args.text,
self.args.cache_dir).reader()
def test_ner(crf, test_sent):
from tokenizer.tokenizer import Tokenizer
token = Tokenizer()
token.run()
arr_featurized_sent = []
postaged_sent = ViPosTagger.postagging(token.predict(test_sent))
print postaged_sent
test_arr = []
for i in xrange(len(postaged_sent[0])):
test_arr.append((postaged_sent[0][i], postaged_sent[1][i]))
print test_arr
featurized_sent = sent2features(test_arr)
arr_featurized_sent.append(featurized_sent)
predict = crf.predict(arr_featurized_sent)
return zip(test_arr, predict[0])
def __init__(self, normalized= True, classes= None, stemmed= True):
if classes is None:
classes = ["positive", "negative", "notr"]
self.x = []
self.y = []
self.tokenizer = Tokenizer()
self.stemmer = TurkishStemmer()
self.word2vec = None
self.cachefile = "data/data" + ("_normalized" if normalized else "") + ("_stemmed" if stemmed else "") + "_" + ("_".join(classes)) + ".pickle"
if os.path.isfile(self.cachefile):
with open(self.cachefile, 'rb') as cache:
self.x, self.y = pickle.load(cache)
else:
for cls in classes:
self._append_data(cls, normalized, stemmed)
with open(self.cachefile, 'wb') as cache:
pickle.dump((self.x, self.y), cache)
def first_stats():
tokenizer = Tokenizer()
tokenizer.run()
question_vocabulary = Vocabulary()
questions = load_questions()
cc = 0
for question in questions:
#print question
if cc % 10 == 0:
print "\r%s" % cc,
cc += 1
sen = tokenizer.predict(question)
sen = sen.lower()
tokens = question_vocabulary.get_sentence_token_ids(sen)
question_list.append(tokens)
print "\n Saving..."
question_vocabulary.save(Q_VOCAB_NAME)
utils.pickle_save(question_list, "question_tokens.dat")
print "Done"
def tokenize(path):
t = Tokenizer(path)
tokens = t.raw_tokenize()
items = []
for token in tokens:
if token.kind.name == "LITERAL":
text = token.spelling
cursor_kind = clang.cindex.CursorKind
kind = token.cursor.kind
if kind == cursor_kind.STRING_LITERAL:
# do extra processing on strings
text = sha256(mangle_text(token.spelling)).hexdigest()[:10]
items.append(text)
if token.kind.name == "COMMENT":
hashed = sha256(mangle_text(token.spelling[2:])).hexdigest()[:10]
items.append(hashed)
return "\n".join(items)
def load(cls,
path,
fields,
tokenizer_lang,
tokenizer_dir,
verbose=True,
max_sent_length=math.inf):
tokenizer = Tokenizer(lang=tokenizer_lang,
dir=tokenizer_dir,
verbose=verbose)
sentences = []
fields = [
field if field is not None else Field(str(i))
for i, field in enumerate(fields)
]
with open(path, 'r') as f:
lines = []
for line in tokenizer.format(tokenizer.predict(f.read())):
line = line.strip()
if not line:
if len(lines) > max_sent_length:
logger.info(
'Discarded sentence longer than max_sent_length:',
len(lines),
file=sys.stderr)
lines = []
continue
sentences.append(Sentence(fields, lines))
lines = []
else:
if not line.startswith('#'):
# append empty columns
line += '\t_' * (len(CoNLL._fields) -
len(line.split('\t')))
lines.append(line)
return cls(fields, sentences)
def predict(self,
X,
part_of_speech=None,
tagger_preprocessed=False,
sentence_level=False):
i = 0
if sentence_level:
results = []
for text in X:
tokenizer = Tokenizer()
sentences = tokenizer.tokenize([text])
sentences = [
" ".join([token[0] for token in sentence])
for sentence in sentences
]
preprocessed_sentences = self.preprocess_texts(
sentences,
part_of_speech=part_of_speech,
tagger_preprocessed=tagger_preprocessed)
X = self.vectorizer.transform(preprocessed_sentences).toarray()
pred = self.nb_model.predict(X)
results.append(int(round(np.mean(pred))))
print(i)
i += 1
return np.array(results)
else:
preprocessed = self.preprocess_texts(
X,
part_of_speech=part_of_speech,
tagger_preprocessed=tagger_preprocessed)
X = self.vectorizer.transform(preprocessed).toarray()
return self.nb_model.predict(X)
print(' tokenizer algorithm : '+str(args.tokenizer))
if args.v is True:
print(' verbose (print debug) : ON')
# output directory
if not os.path.exists(args.p):
os.mkdir(args.p)
# (0) torch settings
torch.manual_seed(args.seed)
torch.cuda.manual_seed(args.seed)
torch.backends.cudnn.deterministic = True
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
# (1) tokenizer setting
tokenizer = Tokenizer(args.tokenizer, '../tokenizer/e2e.model')
# (2) corpus data
random.seed(args.seed)
lex_flag = True
dataset_train = MyDataset(args.corpus.rstrip('/')+'/e2e_train.json',
args.corpus.rstrip('/')+'/e2e_valid.json',
args.corpus.rstrip('/')+'/e2e_test.json',
args.corpus.rstrip('/')+'/e2e-augment/e2e_mr_lex_max_num_token.json',
'train', tokenizer, lex_flag, device)
dataset_valid = MyDataset(args.corpus.rstrip('/')+'/e2e_train.json',
args.corpus.rstrip('/')+'/e2e_valid.json',
args.corpus.rstrip('/')+'/e2e_test.json',
args.corpus.rstrip('/')+'/e2e-augment/e2e_mr_lex_max_num_token.json',
'valid', tokenizer, lex_flag, device)
dataset_test = MyDataset(args.corpus.rstrip('/')+'/e2e_train.json',
# -*- encoding: utf8 -*-
import re
import requests
import unicodedata
from tokenizer.tokenizer import Tokenizer
from sklearn.externals import joblib
import pandas as pd
import os
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.metrics import f1_score
from sklearn.svm import SVC
from pyvi.pyvi import ViTokenizer
from sklearn.metrics import confusion_matrix
tokenizer = Tokenizer()
tokenizer.run()
def load_model(model):
print('loading model ...', model)
if os.path.isfile(model):
return joblib.load(model)
else:
return None
def list_words(mes):
words = mes.lower().split()
return " ".join(words)
def main():
dataset = data_loader.load_text_file("data_2.txt")
tokenizer = Tokenizer()
output = tokenizer.tokenize([dataset])
for sentence in output:
print(sentence)
tf.import_graph_def(restored_graph_def,
input_map=None,
return_elements=None,
name="")
graph = tf.get_default_graph()
doc_ids = graph.get_tensor_by_name('doc_ids:0')
doc_mask = graph.get_tensor_by_name('doc_mask:0')
doc_type = graph.get_tensor_by_name('doc_type:0')
#content = graph.get_tensor_by_name('content:0')
doc_output = graph.get_tensor_by_name('doc/output:0')
doc_max_length = 12
s = Simplifier('tokenizer/zh_mapping.txt')
t = Tokenizer('tokenizer/spiece_all_bpe/spiece.all.bpe.130000.lower.model',
'tokenizer/lg.all.voc', doc_max_length)
count = 0
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
#f = ['a real test', 'b false test']
with open(infile, 'r', encoding='utf-8') as f, open(outfile,
'w',
encoding='utf-8') as fo:
with tf.Session(config=config) as sess:
time = datetime.datetime.now()
for line in f:
simple = s.simplify(line)
tokens = t.tokenize(simple)
accents = run_strip_accents(tokens)
ids = t.token_to_id(accents)
def test_tokenize(self):
tokenizer = Tokenizer(self.args['lang'])
sentences = tokenizer.predict(
'Ha chiamato il dr. Rossi.Vuole salutarti.')
self.assertEqual(len(sentences), 2)
def test_tokenize(self):
tokenizer = Tokenizer(**self.args)
sentences = tokenizer.predict(
'Domani vorrei andare al mare.Speriamo faccia bel tempo.')
self.assertEqual(len(sentences), 2)
def genStats(path, helpers):
t = Tokenizer(path)
tokens = t.raw_tokenize()
# stats
numLines = 0
numWhitespace = 0
numComments = 0
avgIdentLength = 0
numFunctions = 0 # ident followed by (, declarations and calls
numDefines = 0
numMathOps = 0
lenLongestLine = 0
numReturns = 0
# other data
idents = []
text = io.readFile(path)
lastWasIdent = False
# get info from tokens
for token in tokens:
# look for a comment
if token.kind.name == "COMMENT":
numComments += 1
# look for math ops
if token.spelling in [
"+", "-", "*", "/", "|", "&", "+=", "-=", "*=", "/=", ">>=",
"<<=", "++", "--", "~", ">>", "!"
]:
numMathOps += 1
# look for function decs/calls
if lastWasIdent and token.spelling == "(":
numFunctions += 1
# count the number of returns
if token.spelling == "return":
numReturns += 1
# add the identifier to the list, set lastWasIdent
if token.kind.name == "IDENTIFIER":
idents.append(token.spelling)
lastWasIdent = True
else:
lastWasIdent = False
# get average ident length
total = 0.0
for ident in idents:
total += float(len(ident))
avgIdentLength = 0.0
if len(idents) > 0:
avgIdentLenth = total / float(len(idents))
# find the number of defines
defines = re.findall("#\s*define ", text.lower())
numDefines = len(defines)
# find the number of lines
lines = text.split("\n")
if len(lines) == 1:
# ugh, windows
lines = text.split("\r")
numLines = len(lines)
# get the length of the longest line
for line in lines:
if len(line) > lenLongestLine:
lenLongestLine = len(line)
# find the total amount of whitespace
for char in text:
if char in [" ", "\n", "\t", "\r"]:
numWhitespace += 1
# create a dict of results and return
results = {}
results["numLines"] = numLines
results["numWhitespace"] = numWhitespace
results["numComments"] = numComments
results["avgIdentLength"] = avgIdentLength
results["numFunctions"] = numFunctions
results["numDefines"] = numDefines
results["numMathOps"] = numMathOps
results["numReturns"] = numReturns
results["lenLongestLine"] = lenLongestLine
return results
print('startword: ' + startword)
print('** output **')
# MR(LEX) -> TXT_LEX
nlg_txt, attention_txt = NLG_model.convert_nlg(mr_obj, args.search,
lex_flag, startword)
obj_txt = {'txt': nlg_txt}
print('TXT: ' + obj_txt['txt'])
#print(obj_txt)
f = open(args.o, 'w', encoding='utf-8')
json.dump(obj_txt, f, ensure_ascii=False, indent=4, sort_keys=False)
f.close()
print('** done **')
tokenizer = Tokenizer('nltk', '../../tokenizer/e2e.model')
txt = nlg_txt
mr = mr_obj
if read_mr_obj['name'] != '':
txt = txt.replace(read_mr_obj['name'], 'NAME')
mr['name'] = 'NAME'
if read_mr_obj['near'] != '':
txt = txt.replace(read_mr_obj['near'], 'NEAR')
mr['near'] = 'NEAR'
mr_token = tokenizer.mr(mr)
txt_token = tokenizer.txt(txt)
'''
print(mr_token)
print(str(len(mr_token)))
import pickle
from bm25 import BM25
# Special vocabulary symbols - we always put them at the start.
_PAD = b"_PAD"
_GO = b"_GO"
_EOS = b"_EOS"
_UNK = b"_UNK"
_START_VOCAB = [_PAD, _GO, _EOS, _UNK]
PAD_ID = 0
GO_ID = 1
EOS_ID = 2
UNK_ID = 3
default_tokenizer = Tokenizer(_UNK)
bpe_tokenizer = BPETokenizer(
open("/home/martin/projects/subword-nmt/vocab_bpe_merged"), _START_VOCAB)
def create_vocabulary(vocabulary_path,
data_path,
max_vocabulary_size,
dataset_reader,
tokenizer=default_tokenizer,
persist_counts=False):
"""Create vocabulary file (if it does not exist yet) from data file.
Data file is assumed to contain one sentence per line. Each sentence is
tokenized and digits are normalized (if normalize_digits is set).
Vocabulary contains the most-frequent tokens up to max_vocabulary_size.
