def condensify(train):
"""
Takes input either a string or a list of string
Returns a list of all summaries;
For a string returns a list with singleton document
"""
summ_list = []
if isinstance(train,string):
train = [train]
for t in train:
summ=[]
k=0
#corpus = [dictionary.doc2bow(text) for text in texts]
dictionary = corpora.Dictionary([w for w in reuters.sents(t)])
corpus = [dictionary.doc2bow(w) for w in reuters.sents(t)]
matrix = matutils.corpus2csc(corpus)
#print matrix
u,sigma,vt = sparse.linalg.svds(matrix)
(k,l)= vt.shape
while k>=1:
if reuters.sents(t)[vt[k-1].argmax()] not in summ:
summ.append(reuters.sents(t)[vt[k-1].argmax()])
k-=1
v=[]
for s in summ:
v.append(" ".join(s))
summ = "".join(v)
summ_list.append(summ)
return (summ_list)
def import_reuters_flat_pos(ds, silent=False, log=sys.stdout):
"""
Import the brown corpus into `ds`. E.g.
>>> from nathan.core import Dataspace
>>> ds = Dataspace()
>>> %time brown.import_brown(ds, silent=True)
CPU times: user 12min 28s, sys: 536 ms, total: 12min 29s
Wall time: 12min 29s
"""
tagger = nltk.data.load("./models/treebank_brill_aubt/treebank_brill_aubt.pickle")
if not silent:
total = len(reuters.sents())
counter = 0
root_handle = ds.insert("#reuters")
for sent in reuters.sents():
sent = tagger.tag(sent)
norm = [nltk.tuple2str(t) for t in sent]
sen_handle = ds.insert(norm)
ds.link(root_handle, sen_handle)
if not silent:
counter += 1
if (counter % 100 == 0):
print("importing %s of %s sentences..." % (counter, total),
file=log)
def condensify(train):
"""
Takes input either a string or a list of string
Returns a list of all summaries;
For a string returns a list with singleton document
"""
summ_list = []
if isinstance(train, string):
train = [train]
for t in train:
summ = []
k = 0
#corpus = [dictionary.doc2bow(text) for text in texts]
dictionary = corpora.Dictionary([w for w in reuters.sents(t)])
corpus = [dictionary.doc2bow(w) for w in reuters.sents(t)]
matrix = matutils.corpus2csc(corpus)
#print matrix
u, sigma, vt = sparse.linalg.svds(matrix)
(k, l) = vt.shape
while k >= 1:
if reuters.sents(t)[vt[k - 1].argmax()] not in summ:
summ.append(reuters.sents(t)[vt[k - 1].argmax()])
k -= 1
v = []
for s in summ:
v.append(" ".join(s))
summ = "".join(v)
summ_list.append(summ)
return (summ_list)
def getSummaries(cfs, number_of_summaries):
for n in range(number_of_summaries):
# get a random article from the corpus
article = random.choice(reuters.fileids())
# make sure the article is of the apropriate length
# I decided at least 5 sentences!
while len(reuters.sents(article)) < 5:
article = random.choice(reuters.fileids())
length = len(reuters.sents(article)) // 2
summary_sentences = cfs.summarize(article, length)
print_summary(summary_sentences)
def load_reuters(self):
sents = reuters.sents()
print "Done loading reuters, cleaning..."
# clean, etc...
data = []
for sentence in sents:
x = []
for word in sentence:
if word in self.punctset:
continue
w = word.lower().strip(self.punctuation)
ind = self.word2id.get(w, -1)
if ind < 0:
ind = len(self.word2id)
self.word2id[w] = ind
x.append(ind)
data.append(x)
print "Done cleaning reuters, vectorizing..."
X = sparse.lil_matrix((
len(self.word2id),
len(data),
))
for (j, dat) in enumerate(data):
for i in dat:
X[i, j] = 1
return X
def language_mode():
# 读取语料库 from NLTK
categories = reuters.categories()
corpus = reuters.sents(categories=categories)
print(f'top 3 corpus is:\n {corpus[:3]}')
term_count = {}
bigram_count = {}
for doc in corpus:
doc = ['<s>'] + doc
for i in range(0, len(doc) - 1):
#bigram :[i,i +1]
term = doc[i]
bigram = doc[i:i + 2]
if term in term_count:
term_count[term] += 1
else:
term_count[term] = 1
bigram = ' '.join(bigram)
if bigram in bigram_count:
bigram_count[bigram] += 1
else:
bigram_count[bigram] = 1
print(f'term_count length is: \n {len(term_count)}'
) # {'<s>': 54716, 'ASIAN': 12, 'EXPORTERS': 46, 'FEAR'
print(
f'bigram_count length is: \n {len(bigram_count)}'
) #{'<s> ASIAN': 4, 'ASIAN EXPORTERS': 1, 'EXPORTERS FEAR': 1, 'FEAR DAMAGE': 1,
return term_count, bigram_count
def reuters_to_df(set_name, label_to_idx):
data = [x for x in reuters.fileids() if set_name in x]
# collect all data to create df from
all_texts = [
" ".join([" ".join(sen) for sen in reuters.sents(doc_id)])
for doc_id in data
]
all_labels = np.zeros((len(all_texts), len(label_to_idx)))
all_label_indices = [[
label_to_idx[lab] for lab in reuters.categories(doc_id)
] for doc_id in data]
for i, labs in enumerate(all_label_indices):
# binary encode the labels
all_labels[i][labs] = 1
all_labels = all_labels.astype(int)
# all_labels[all_label_indices] = 1
cols = ["text"]
label_cols = ["topic_{}".format(lab) for lab in reuters.categories()]
cols.extend(label_cols)
# create df and set values
df = pd.DataFrame(columns=cols)
df["text"] = all_texts
df[label_cols] = all_labels
return df
def import_reuters_files(ds, silent=False, log=sys.stdout):
"""
Import the brown corpus into `ds`. E.g.
>>> from nathan.core import Dataspace
>>> ds = Dataspace()
>>> %time brown.import_brown(ds, silent=True)
CPU times: user 12min 28s, sys: 536 ms, total: 12min 29s
Wall time: 12min 29s
"""
if not silent:
total = len(reuters.fileids())
counter = 0
root_handle = ds.insert("#reuters")
for fileid in reuters.fileids():
tags = ["@%s" % category for category in reuters.categories(fileid)]
file_handle = ds.insert(["#%s" % fileid] + tags)
ds.link(root_handle, file_handle)
for sent in reuters.sents(fileid):
norm = [word.lower() for word in sent]
sen_handle = ds.insert(norm)
ds.link(file_handle, sen_handle)
if not silent:
counter += 1
if (counter % 10 == 0):
print("importing %s of %s files..." % (counter, total),
file=log)
def filesForEval(cfs):
#the number of sentences in each summary to be generated
length = 5
#the number of summaries to be written into the file
number_of_summaries = 10
#create the file containing the summaries
summFile = open('random_order_summaries_2.txt', 'w')
#create the file which will be the key
keyFile = open('key_2.txt', 'w')
#generate the appropriate number of summaries
for n in range(number_of_summaries):
#get a random article from the corpus
article = random.choice(reuters.fileids())
#make sure the article is of the apropriate length
# I decided at least twice the length of the summary in this case.
while len(reuters.sents(article)) < length * 4:
article = random.choice(reuters.fileids())
#print info about the article into the summary-containing document
summFile.write('Article #' + str(n) + '\n')
summFile.write('\nfileid: ' + article + '\n\n')
#print info about the article into the key document
keyFile.write('Article #' + str(n) + '\n')
keyFile.write('fileid: ' + article + '\n\n')
#get a list of sentences that is the summary generated by our algorithm
summSents = cfs.summarize(article, length)
#insert a marker to make sure we remember it is the summary
summSents.insert(0, 'summ')
#get a list of sentences that were randomly ordered
randSents = getRandom(article, length)
#insert a marker to make sure we remember it is the random selection
randSents.insert(0, 'rand')
#mix up the ordering
summs = [summSents, randSents]
random.shuffle(summs)
#Write the summaries into the file and write the key, in a semi-nice format
for summ in summs:
for i, sentence in enumerate(summ):
if i == 0:
keyFile.write(sentence)
else:
for word in sentence:
summFile.write(word + ' ')
summFile.write('\n')
summFile.write('\n')
keyFile.write(' ')
summFile.write('\n')
keyFile.write('\n')
summFile.close()
keyFile.close()
def __init__(self):
training_files = [
fileid for fileid in reuters.fileids()
if fileid.startswith('training')
]
super(ReutersTrainingCorpus,
self).__init__(reuters.sents(training_files))
def create_raw_data_for_classifier(start_pt, end_pt):
# printing process id
#print("ID of process running : {}".format(os.getpid()))
#df_for_raw = pd.DataFrame(columns=['sentences','polarity'])
pos = 0
neg = 0
polarity_list = []
sentncs_list = []
#for i in range(len(reuters.sents())):
for i in range(start_pt, end_pt):
sentncs = " ".join(reuters.sents()[i])
#print("sentncs = ", sentncs)
blob = TextBlob(sentncs)
sentncs_list.append(sentncs)
if blob.sentiment.polarity > 0:
polarity_list.append('pos')
pos = pos + 1
elif blob.sentiment.polarity < 0:
polarity_list.append('neg')
neg = neg + 1
raw_data = list(zip(sentncs_list, polarity_list))
#print(reutersDf.tail(10))
print("raw_data len = ", len(raw_data))
print("Total pos = ", pos, " Total Neg =", neg)
print(raw_data[0])
raw_data = list(zip(sentncs_list, polarity_list))
#print(reutersDf.tail(10))
print("raw_data len = ", len(raw_data))
print("Total pos = ", pos, " Total Neg =", neg)
print(raw_data[0])
return raw_data
def reuters_idf_dict(current_docs, file_name, order=2):
"""
"""
idf_file = file_name + ".idf"
dict_idf = {}
if os.path.exists(idf_file):
with open(idf_file, 'r', encoding='utf-8') as f:
for line in f:
values = line.split("\t")
# print(values)
dict_idf[tuple(values[0].split())] = float(values[1])
dict_idf = make_concept_idf_dict(current_docs, dict_idf,
len(reuters.fileids()))
return dict_idf
else:
logger.info("Process reuters idf.")
l_docs = []
for fileid in reuters.fileids():
l_docs.append(reuters.sents(fileids=[fileid]))
dict_idf = make_concept_idf_dict(l_docs, order=order)
with open(idf_file, 'w', encoding='utf-8') as f:
for concept in dict_idf.keys():
f.write(' '.join(concept) + '\t' + str(dict_idf[concept]) +
'\n')
dict_idf = make_concept_idf_dict(current_docs, dict_idf,
len(reuters.fileids()))
return dict_idf
def __init__(self, vocab_path: os.PathLike, spell_error_path: os.PathLike):
# 构建词典库
with open(vocab_path) as f:
self.vocab = {line.strip() for line in f}
self.vocab_size = len(self.vocab)
# 获取语料库, 构建语言模型
categories = reuters.categories()
corpus = reuters.sents(categories=categories)
self.unigram_count, self.bigram_count = defaultdict(int), defaultdict(
int)
for doc in corpus:
doc = ['<s>'] + doc
for i in range(1, len(doc)):
self.unigram_count[doc[i]] += 1
self.bigram_count[(doc[i - 1], doc[i])] += 1
# 统计拼写错误概率 P(mistake|correct)
self.channel_prob = defaultdict(dict)
with open(spell_error_path) as f:
for line in f:
temp = line.split(':')
correct = temp[0].strip()
mistakes = [m.strip() for m in temp[1].strip().split(',')]
for m in mistakes:
self.channel_prob[correct][m] = 1. / len(mistakes)
def initCount2():
corpus_raw_text = reuters.sents(categories=reuters.categories())
gram_count = {}
count = [0, 0, 0]
for sents in corpus_raw_text:
sents = ['<s>'] + sents + ['</s>']
# remove string.punctuation
for words in sents[::]: # use [::] to remove the continuous ';' ';'
if (words in [
'\'\'', '``', ',', '--', ';', ':', '(', ')', '&', '\'',
'!', '?', '.'
]):
sents.remove(words)
# count the n-gram
for n in range(1, 3): # only compute 1/2/3-gram
if (len(sents) <= n): # 'This sentence is too short!'
continue
else:
for i in range(n, len(sents) + 1):
gram = sents[i - n:i] # ['richer', 'fuller', 'life']
key = ' '.join(gram) # richer fuller life
count[n] = count[n] + 1
if (key in gram_count): # use dict's hash
gram_count[key] += 1
else:
gram_count[key] = 1
with open("Count.pk", "wb") as fCount:
pickle.dump([gram_count], fCount)
return gram_count, count[0], count[1], count[2]
def getreuters(): # don't tokenize from nltk.corpus import reuters reuterslist=[] for article in reuters.fileids(): reuterslist.append(reuters.sents(article)) return reuterslist
def get_tokenized_sentences(dataset):
if dataset == 'brown_corpus':
return list(brown.sents())
elif dataset == 'reuters_corpus':
return list(reuters.sents())
elif dataset == 'gatsby':
with open('./data/gatsby.txt', 'r') as f:
text = '\n'.join(f.readlines())
tok_sent = [word_tokenize(t) for t in sent_tokenize(text)]
return tok_sent
elif dataset == 'RACE_corpus':
df_1 = pd.read_csv('./data/middle_combined.csv')
df_2 = pd.read_csv('./data/high_combined.csv')
text = '\n'.join(list(df_1['text']) + list(df_2['text']))
# This dataset appears to have an issue with period spacing
text = text.replace(".", ". ")
tok_sent = [word_tokenize(t) for t in sent_tokenize(text)]
return tok_sent
elif dataset in ('news_small', 'news_large'):
df = pd.read_csv(f'./data/all_the_{dataset}.csv')
text = '\n'.join(list(df['content']))
text = text.replace(" ", " ")
text = text.replace(" ", " ")
tok_sent = [word_tokenize(t) for t in sent_tokenize(text)]
return tok_sent
elif dataset.startswith("books"):
# Get all books
if dataset == 'books':
text = ""
dif = ["middle", "high", "college"]
for d in dif:
files = glob.glob(f"./data/books/{d}/*.txt")
for file in files:
with open(file, 'r') as f:
text += "\n".join(f.readlines()) + "\n"
# Get all books of specified difficulty
else:
difficulty = dataset.split("_")[1]
files = glob.glob(f"./data/books/{difficulty}/*.txt")
text = ""
for file in files:
with open(file, 'r') as f:
text += "\n".join(f.readlines()) + "\n"
# Underscores are used to indicate italics here and should be dropped.
text = text.replace("_", "")
tok_sent = [word_tokenize(t) for t in sent_tokenize(text)]
return tok_sent
# Assume .txt document
else:
with open(dataset, 'r') as f:
text = "\n".join(f.readlines())
tok_sent = [word_tokenize(t) for t in sent_tokenize(text)]
return tok_sent
def load_reuters():
from nltk.corpus import reuters
text = reuters.sents()
text = [[word.lower() for word in sentence] for sentence in text]
vocab = Vocab.build(text,
reserved_tokens=[BOS_TOKEN, EOS_TOKEN, PAD_TOKEN])
corpus = [vocab.convert_tokens_to_idx(sentence) for sentence in text]
return corpus, vocab
def q4(query):
print("\n" + "~"*10 + " Q4 " + "~"*10)
# retrieve seventh sentence and join into a string
# also remove punctuation from it since they are separated out
pick = " ".join( remove_punc(word) for word in reuters.sents()[7] if word not in punc)
result = (pick, jaccard(query, pick))
print(result)
def q1():
global docs
# retrieve sentences from three genres
docs = list(reuters.sents(categories='bop'))
docs = docs + (reuters.sents(categories='cocoa'))
docs = docs + (reuters.sents(categories='zinc'))
#print(docs)
# convert list of lists into list of strings
doc = []
for sent in docs:
# case fold each sentence
doc.append(" ".join([
word.lower() for word in sent
if word not in set(string.punctuation)
]))
docs = doc
def get_corpus():
"""To get nltk corpora."""
from nltk.corpus import brown
from nltk.corpus import reuters
corpus = add_sent_marker(brown.sents())
corpus = corpus + add_sent_marker(reuters.sents())
corpus = add_sent_marker(corpus)
return corpus
def train_skip(CBOW_skip=0, embed_size=100, window=5, min_count=5, epochs=5, workers=1):
corpus_name = 'Brown'
corpus = itertools.chain(reuters.sents(), brown.sents(), gutenberg.sents())
corpus = list(corpus)
model = Word2Vec(corpus,sg=CBOW_skip, size=embed_size, window=window, min_count=min_count, workers=workers)
logging.warning("[1]"+",".join([corpus_name,str(CBOW_skip),str(embed_size),str(window)]))
model.train(corpus, total_examples=len(corpus), epochs=epochs)
model = gensim_to_embed(model)
return wordsim_eval(model)
def get_reuters_token_list_by_sentence(num_doc=100):
""" Get a test data from reuters corpus.
Stopwords will be included to see how HMM_LDA works with these stopwords.
Parameters
----------
num_doc: int
number of documents to be returned
max_voca
maximum number of vocabulary size for the returned corpus
Returns
-------
voca: ndarray
vocabulary
corpus: list
nested list of
"""
file_list = reuters.fileids()
corpus = [reuters.sents(file_list[i]) for i in xrange(num_doc)]
valid_voca = set(w.lower() for w in nltk.corpus.words.words())
stop = stopwords.words('english')
valid_voca = valid_voca.union(stop)
tmp_corpus = list()
voca_dic = dict()
voca = list()
for doc in corpus:
tmp_doc = list()
for sent in doc:
tmp_sent = list()
for word in sent:
if word in valid_voca:
tmp_sent.append(word)
if word not in voca_dic:
voca_dic[word] = len(voca_dic)
voca.append(word)
if len(tmp_sent) > 0:
tmp_doc.append(tmp_sent)
if len(tmp_doc) > 0:
tmp_corpus.append(tmp_doc)
# convert token list to word index list
corpus = list()
for doc in tmp_corpus:
new_doc = list()
for sent in doc:
new_sent = list()
for word in sent:
new_sent.append(voca_dic[word])
new_doc.append(new_sent)
corpus.append(new_doc)
return np.array(voca), corpus
def get_reuters_token_list_by_sentence(num_doc=100):
""" Get a test data from reuters corpus.
Stopwords will be included to see how HMM_LDA works with these stopwords.
Parameters
----------
num_doc: int
number of documents to be returned
max_voca
maximum number of vocabulary size for the returned corpus
Returns
-------
voca: ndarray
vocabulary
corpus: list
nested list of
"""
file_list = reuters.fileids()
corpus = [reuters.sents(file_list[i]) for i in xrange(num_doc)]
valid_voca = set(w.lower() for w in nltk.corpus.words.words())
stop = stopwords.words('english')
valid_voca = valid_voca.union(stop)
tmp_corpus = list()
voca_dic = dict()
voca = list()
for doc in corpus:
tmp_doc = list()
for sent in doc:
tmp_sent = list()
for word in sent:
if word in valid_voca:
tmp_sent.append(word)
if word not in voca_dic:
voca_dic[word] = len(voca_dic)
voca.append(word)
if len(tmp_sent) > 0:
tmp_doc.append(tmp_sent)
if len(tmp_doc) > 0:
tmp_corpus.append(tmp_doc)
# convert token list to word index list
corpus = list()
for doc in tmp_corpus:
new_doc = list()
for sent in doc:
new_sent = list()
for word in sent:
new_sent.append(voca_dic[word])
new_doc.append(new_sent)
corpus.append(new_doc)
return np.array(voca), corpus
def create_model_from_NLTK():
filepath = "nltkcorpus.txt"
if isfile(filepath):
return create_model(filepath= filepath, save=False)
else:
from nltk.corpus import reuters, brown, gutenberg
sents = reuters.sents() + brown.sents()
for gsents in [gutenberg.sents(fid) for fid in gutenberg.fileids()]:
sents += gsents
return create_model(sentences=sents, savename=filepath)
def create_corpus():
"""Removing punctuations from the sentences of corpus."""
# Here, I am taking a text from rueters, webtext and brown corpus here.
rr_corpus = reuters.sents() + webtext.sents() + brown.sents()
punctuations = [p for p in string.punctuation]
cleaned_corpus = []
for idx in range(len(rr_corpus)):
cleaned_corpus.append(
[w for w in rr_corpus[idx] if w not in punctuations])
return cleaned_corpus
def _load_reuters_docs():
test_docs = []
train_docs = []
i = 0
for fileid in reuters.fileids():
i += 1
if 'test' in fileid:
# test_docs.append((reuters.words(fileid), reuters.sents(fileid)))
test_docs.append(
TokenizedDoc(reuters.words(fileid), reuters.sents(fileid),
reuters.categories(fileid)))
elif 'training' in fileid:
# train_docs.append((reuters.words(fileid), reuters.words(fileid)))
train_docs.append(
TokenizedDoc(reuters.words(fileid), reuters.sents(fileid),
reuters.categories(fileid)))
else:
print(
"Document not recognized as part of training-set or test-set while extracting the Reuters Corpus"
)
return train_docs, test_docs
def load_reuters_corpus() -> List[str]:
nltk.download('reuters')
sentences = list(
filter(
lambda sent: (len(sent) <= 30) and
(len(sent) >= 3) and any(map(lambda word: word.isalpha(), sent))
and len(list(filter(lambda word2: word2.isupper(), sent))) <
(len(sent) // 4), reuters.sents()))
mdetok = TreebankWordDetokenizer()
return list(
map(
lambda sent: mdetok.detokenize(
(' '.join(sent).replace('``', '"').replace("''", '"').replace(
'`', "'")).split()), sentences))
def process_reuters():
print 'reuters'
from nltk.corpus import reuters
count = 0
word = 'bank'
sen1 = 'depository_financial_institution.n.01'
sen2 = 'bank.n.01'
file_name = 'data/bank_reuters_tmp.txt'
for f in reuters.fileids():
sents = reuters.sents(f)
for i in range(len(sents)):
sent = sents[i]
if (word in sent):
appendToFile(file_name, sentToStr(sent, '0'))
count = count + 1
print count
def pre_process():
raw_sentences = reuters.sents(reuters.fileids())
tokenizer = Tokenizer()
tokenizer.fit_on_texts(raw_sentences)
count_thresh = 5
low_count_words = [
w for w, c in tokenizer.word_counts.items() if c < count_thresh
]
for w in low_count_words:
del tokenizer.word_index[w]
del tokenizer.word_docs[w]
del tokenizer.word_counts[w]
word_index_dict = tokenizer.word_index
index_word_dict = {word_index_dict[word]: word for word in word_index_dict}
sentences_word_index = tokenizer.texts_to_sequences(raw_sentences)
return sentences_word_index, word_index_dict, index_word_dict
def get_default_sentences() -> list:
nltk.download('brown')
brown_tokenized_sentences = brown.sents()
brown_sentences = detok_sentences(brown_tokenized_sentences)
nltk.download('gutenberg')
nltk.download('punkt')
gutenberg_tokenized_sentences = gutenberg.sents()
gutenberg_sentences = detok_sentences(gutenberg_tokenized_sentences)
nltk.download('reuters')
reuters_tokenized_sentences = reuters.sents()
reuters_sentences = detok_sentences(reuters_tokenized_sentences)
nltk.download('webtext')
webtext_tokenized_sentences = webtext.sents()
webtext_sentences = detok_sentences(webtext_tokenized_sentences)
nltk.download('inaugural')
inaugural_tokenized_sentences = inaugural.sents()
inaugural_sentences = detok_sentences(inaugural_tokenized_sentences)
return brown_sentences + gutenberg_sentences + reuters_sentences + webtext_sentences + inaugural_sentences
def N_gramprediction(self, input_text):
from nltk.corpus import reuters
from nltk import bigrams, trigrams
from collections import Counter, defaultdict
import random
# Create a placeholder for model
model = defaultdict(lambda: defaultdict(lambda: 0))
# Count frequency of co-occurance
for sentence in reuters.sents():
for w1, w2, w3 in trigrams(sentence, pad_right=True,
pad_left=True):
model[(w1, w2)][w3] += 1
# Let's transform the counts to probabilities
for w1_w2 in model:
total_count = float(sum(model[w1_w2].values()))
for w3 in model[w1_w2]:
model[w1_w2][w3] /= total_count
# starting word
text1 = str(input_text)
text = list(text1.split(' '))
sentence_finished = False
print(text1)
while not sentence_finished:
# select a random probability threshold
r = random.random()
accumulator = .0
for word in model[tuple(text[-2:])].keys():
accumulator += model[tuple(text[-2:])][word]
# select words that are above the probability threshold
if accumulator >= r:
text.append(word)
break
if text[-2:] == [None, None]:
sentence_finished = True
return (' '.join([t for t in text if t]))
def getRandom(article, length):
allSents = list(enumerate(list(reuters.sents(article))))
print(len(allSents))
sentences = []
sentences.append(allSents[0])
allSents.__delitem__(0)
for n in range(length):
sent = random.choice(allSents)
sentences.append(sent)
allSents.remove(sent)
sentences.sort()
sentList = []
for sent in sentences:
sentence = sent[1]
sentList.append(sentence)
return sentList
def reuters_idf_dict(current_docs, file_name, order=2):
"""
"""
idf_file = file_name + ".idf"
dict_idf = {}
if os.path.exists(idf_file):
with open(idf_file, 'r') as f:
for line in f:
values = line.split("\t")
dict_idf[tuple(values[0].split())] = values[1]
return dict_idf
else:
l_docs = []
for doc in current_docs:
l_docs.append(doc)
for fileid in reuters.fileids():
l_docs.append(reuters.sents(fileids=[fileid]))
dict_idf = make_concept_idf_dict(l_docs, order)
with open(idf_file, 'w') as f:
for concept in dict_idf.keys():
f.write(' '.join(concept) + '\t' + str(dict_idf[concept]) +
'\n')
return dict_idf
datafile = open(datapath, "r")
datalines = []
for i in range(1000):
dataline = datafile.readline().split('\t')
datalines.append(dataline)
datafile.close()
end = time.time()
print("time of loading data", end - start)
# begin correction
start = time.time()
n = 100
fres = open("result.txt", "w")
corpus_raw_text = reuters.sents(categories=reuters.categories())
corpus_text = []
for sents in corpus_raw_text:
sents = ['<s>'] + sents + ['</s>']
# remove string.punctuation
for words in sents[::]: # use [::] to remove the continuous ';' ';'
if (words in [
'\'\'', '``', ',', '--', ';', ':', '(', ')', '&', '\'', '!',
'?', '.'
]):
sents.remove(words)
corpus_text.extend(sents)
vocab_corpus = {}.fromkeys(corpus_text).keys()
vocab_corpus = list(vocab_corpus)
vocabCount = len(vocab_corpus)
corpus_str = ' '.join(corpus_text)
def __init__(self, language):
self.language = language
# from 'Multilingual and Cross-Lingual Complex Word Identification' (Yimam et. al, 2017)
if language == 'english':
self.avg_word_length = 5.3
# from Beker, Henry; Piper, Fred. Cipher Systems: The Protection of Communications.
self.char_frequency = {
'a': 8.167,
'b': 1.492,
'c': 2.782,
'd': 4.253,
'e': 12.702,
'f': 2.228,
'g': 2.015,
'h': 6.094,
'i': 6.966,
'j': 0.153,
'k': 0.772,
'l': 4.025,
'm': 2.406,
'n': 6.749,
'o': 7.507,
'p': 1.929,
'q': 0.095,
'r': 5.987,
's': 6.327,
't': 9.056,
'u': 2.758,
'v': 0.978,
'w': 2.360,
'x': 0.150,
'y': 1.974,
'z': 0.074
}
self.dic = pyphen.Pyphen(lang='en')
self.reuters = reuters.words()
self.unigram_counts = Counter(self.reuters)
bigrams = []
for sent in reuters.sents():
bigrams.extend(
nltk.bigrams(sent, pad_left=True, pad_right=True))
self.bigram_counts = Counter(bigrams)
else: # spanish
self.avg_word_length = 6.2
# self.char_frequency = {'a': 12.525,'b': 2.215,'c': 4.139,'d': 5.860,'e': 13.681,
# 'f': 0.692,'g': 1.768,'h': 0.703,'i': 6.247,'j': 0.443,
# 'k': 0.011,'l': 4.967,'m': 3.157,'n': 6.71,'o': 8.683,
# 'p': 2.510, 'q': 0.877,'r': 6.871,'s': 7.977,'t': 4.632,
# 'u': 3.927, 'v': 1.138,'w': 0.017,'x': 0.215,'y': 1.008,
# 'z': 0.517,'á': 0.502, 'é': 0.433, 'í': 0.725, 'ñ': 0.311,
# 'ó': 0.827, 'ú': 0.168, 'ü': 0.012}
# self.dic = pyphen.Pyphen(lang='es')
self.cess = cess.words()
self.unigram_counts = Counter(self.cess)
bigrams = []
for sent in cess.sents():
bigrams.extend(
nltk.bigrams(sent, pad_left=True, pad_right=True))
self.bigram_counts = Counter(bigrams)
# self.clf = svm.SVC()
# self.model = LogisticRegression()
self.model = svm.SVC(gamma=5)
#!/usr/local/bin/python
import nltk, re, pprint
import triple.py
from nltk.corpus import reuters
from nltk.sem import relextract,extract_rels,rtuple
grammar = "Relation: {<DT>?<JJ>*<NN><V.*><NN>}"
cp = nltk.RegexpParser(grammar)
s = [nltk.pos_tag(s) for s in reuters.sents()[:30]]
#print sentence
#print cp.parse(sentence)
#nltk.ne_chunk
brown = nltk.corpus.brown #
for sent in s:
tree = cp.parse(sent)
for subtree in tree.subtrees():
if subtree.label() == 'Relation':
print(tree)
print(subtree.leaves())
'''#
IN = re.compile(r'.*\bof\b.*')
for i,sent in enumerate(s):
sent = nltk.ne_chunk(sent)
rels = extract_rels('PERSON','ORGANIZATION',doc=sent,corpus='ace',pattern=IN,window=7)
self.logger.info("returning top terms")
return [term for term, ig in term_ig[:k]]
def top_common_words(self, k):
self.logger.info("calculating top %d of %d word terms according to frequency", k, len(self._total_freq))
# terms = self.all_terms()
# terms_freq = [(term, sum((term.frequency(doc) for doc in self._documents))) for term in terms]
terms_freq = sorted(self._total_freq.items(), key=lambda x: x[1], reverse=True)
self.logger.info("returning top %d word terms according to frequency", k)
return [term for term, freq in terms_freq[:k]]
if __name__ == '__main__':
training_fileids = fileids = filter(lambda x: "training" in x, reuters.fileids())
documents = reuters.sents(training_fileids)
# dict = set(reuters.words(training_fileids))
# print documents[0]
# print " ".join(documents[0])
# print WordTerm("in").frequency(documents[0])
print 'Checking Vectorizer'
documents = get_document_objects(documents)
w = WordTermExtractor(documents, None)
print documents[0].get_freq('BAHIA')
print documents[0].get_freq('bahia')
print ProjectParams.terms_matrix.total_freq
print "Processed {0} sentences\r".format(processed_count),
print "Current Structure total: {0}".format(len(sentences))
print "Adding abc sentence structures ({0})...".format(len(abc.sents()))
for sentence in abc.sents():
processed_count += 1
try:
blob = TextBlob(filter(lambda x: x in string.printable, " ".join(sentence)), pos_tagger=PerceptronTagger())
tags = tuple([tag[1] for tag in blob.tags])
sentences.add(tags)
except:
print "\r",
print "Processed {0} sentences\r".format(processed_count),
print "Current Structure total: {0}".format(len(sentences))
print "Adding reuters sentence structures ({0})...".format(len(reuters.sents()))
for sentence in reuters.sents():
processed_count += 1
try:
blob = TextBlob(filter(lambda x: x in string.printable, " ".join(sentence)), pos_tagger=PerceptronTagger())
tags = tuple([tag[1] for tag in blob.tags])
sentences.add(tags)
except:
print "\r",
print "Processed {0} sentences\r".format(processed_count),
print "Current Structure total: {0}".format(len(sentences))
print "Adding brown sentence structures ({0})...".format(len(brown.sents()))
for sentence in brown.sents():
processed_count += 1
try:
"""
There are three options to train the true caser:
1) Use the sentences in NLTK
2) Use the train.txt file. Each line must contain a single sentence. Use a large corpus, for example Wikipedia
3) Use Bigrams + Trigrams count from the website http://www.ngrams.info/download_coca.asp
The more training data, the better the results
"""
# :: Option 1: Train it based on NLTK corpus ::
print "Update from NLTK Corpus"
NLTKCorpus = brown.sents()+reuters.sents()+nltk.corpus.semcor.sents()+nltk.corpus.conll2000.sents()+nltk.corpus.state_union.sents()
updateDistributionsFromSentences(NLTKCorpus, wordCasingLookup, uniDist, backwardBiDist, forwardBiDist, trigramDist)
# :: Option 2: Train it based the train.txt file ::
""" #Uncomment, if you want to train from train.txt
print "Update from train.txt file"
sentences = []
for line in open('train.txt'):
sentences.append(line.strip())
tokens = [nltk.word_tokenize(sentence) for sentence in sentences]
updateDistributionsFromSentences(tokens, wordCasingLookup, uniDist, backwardBiDist, forwardBiDist, trigramDist)
"""
# :: Option 3: Train it based ngrams tables from http://www.ngrams.info/download_coca.asp ::
""" #Uncomment, if you want to train from train.txt
# def terminals(self, term, document):
# return (
# self.bool(term, document), self.tf(term, document), self.tf_idf(term, document), self.tf_ig(term, document),
# self.tf_chi(term, document), self.tf_rf(term, document))
def raw_terminals(self, term, document):
return (self.bool(term, document), self.tf(term, document), self.max_prob_term_and_category(term, document),
self.max_prob_term_not_category(term, document), self.avg_prob_term_category(term, document),
self.avg_prob_term_not_category(term, document), self.first_occurrence_perc(term, document))
if __name__ == '__main__':
training_fileids = fileids = filter(lambda fileid: "training" in fileid and len(reuters.categories(fileid)) == 1,
reuters.fileids())
documents = [sum(reuters.sents(fid), []) for fid in training_fileids]
doc = documents[0]
term = terminals.WordTerm("in")
docs_categories = [reuters.categories(fid)[0] for fid in training_fileids]
print docs_categories
print doc
fe = TWSCalculator(documents, docs_categories)
print "tf =", fe.tf(term, doc), "idf =", fe.idf(term), "tf-idf =", fe.tf_idf(term, doc)
term = terminals.WordTerm("in")
print 'TF-CHI: ', fe.tf_chi(term, doc)
print 'TF-CHI: ', fe.tf_chi(term, doc)
print 'TF-IG: ', fe.tf_ig(term, doc)
print 'TF-IG: ', fe.tf_ig(term, doc)
from sklearn.multiclass import OneVsRestClassifier
from sklearn.naive_bayes import MultinomialNB
from features import TWSCalculator
from readers import NewsgroupsReader
from terminals import get_document_objects, WordTermExtractor, WordTerm
from terms_lists.ng20_ig import ng_20_ig500
from terms_lists.r8_ig import r_eight_terms
__author__ = 'itay'
if __name__ == '__main__':
cats_limiter = categories = ['earn', 'acq', 'crude', 'trade', 'money-fx', 'interest', 'money-supply',
'ship'] # top 8
training_fileids = fileids = filter(lambda fileid: "training" in fileid and len(reuters.categories(fileid)) == 1,
reuters.fileids(cats_limiter))
training_documents = [" ".join(sum(reuters.sents(fid), [])) for fid in training_fileids]
training_docs_categories = [reuters.categories(fid)[0] for fid in training_fileids]
map(lambda x: x.lower, training_documents)
# training_documents_objects = get_document_objects(training_documents, training_docs_categories)
# training_documents_objects = NewsgroupsReader(False).get_training()
# training_documents = [d.doc for d in training_documents_objects]
# training_docs_categories = [d.category for d in training_documents_objects]
#top IG r8:
# words = ng_20_ig500
# tws_calculator = TWSCalculator(training_documents_objects, training_docs_categories)
# word_term_extractor = WordTermExtractor(training_documents_objects, tws_calculator)
#
# top_terms = word_term_extractor.top_common_words(500)
def __init__(self):
test_files = [fileid for fileid in reuters.fileids() if fileid.startswith('test')]
super(ReutersTestCorpus, self).__init__(reuters.sents(test_files))
#!/usr/bin/python #coding:utf-8 # 2013/02/08 from nltk.corpus import reuters import nltk # nltk.pos_tag() のタグの種類を調べる sents=reuters.sents() tags = set( tag for sent in sents[:5000] for word,tag in nltk.pos_tag(sent) ) print tags # set(['PRP$', 'VBG', 'VBD', '``', 'VBN', ',', "''", 'VBP', 'WDT', 'JJ', 'WP', 'VBZ', 'DT', 'RP', 'NN', 'POS', '.', 'TO', 'PRP', 'RB', ':', 'NNS', 'NNP', 'VB', 'WRB', 'CC', 'LS', 'PDT', 'RBS', 'RBR', 'CD', '-NONE-', 'EX', 'IN', 'WP$', 'MD', 'NNPS', 'JJS', 'JJR']) # VB,VBG,VBD,VBN,VBP,VBZ,JJ,JJS,JJR,NN,NNS,NNP,NNPS # VB : 動詞,基本形 # VBD : 過去形 # VBG : 動名詞,現在分詞 # VBN : 過去分詞 # VBP : 非三単現 # VBZ : 三単現 # JJ : 形容詞 # JJR : 比較級 # JJS : 最上級 # NN : 単数名詞 # NNS : 複数名詞 # NNP : 単数固有名詞 # NNPS : 複数固有名詞
def __init__(self):
training_files = [fileid for fileid in reuters.fileids() if fileid.startswith('training')]
super(ReutersTrainingCorpus, self).__init__(reuters.sents(training_files))
