forked from zll17/Neural_Topic_Models
-
Notifications
You must be signed in to change notification settings - Fork 0
/
dataset.py
149 lines (136 loc) · 6.29 KB
/
dataset.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
#!/usr/bin/env python
# coding: utf-8
import os
import time
import numpy as np
import pandas as pd
import gensim
import pickle
import random
import torch
from tqdm import tqdm
from tokenization import Tokenizer
from torch.utils.data import Dataset,DataLoader
from collections import Counter
from gensim.corpora import Dictionary
from gensim.models import TfidfModel
from collections import Counter
class DocDataset(Dataset):
def __init__(self,taskname,txtPath=None,tokenizer=None,stopwords=None,no_below=5,no_above=0.1,hasLable=False,rebuild=False,use_tfidf=False):
cwd = os.getcwd()
txtPath = os.path.join(cwd,'data',f'{taskname}_lines.txt') if txtPath==None else txtPath
tmpDir = os.path.join(cwd,'data',taskname)
self.txtLines = [line.strip('\n') for line in open(txtPath,'r',encoding='utf-8')]
self.dictionary = None
self.bows,self.docs = None,None
self.use_tfidf = use_tfidf
self.tfidf,self.tfidf_model = None,None
if not os.path.exists(tmpDir):
os.mkdir(tmpDir)
if not rebuild and os.path.exists(os.path.join(tmpDir,'corpus.mm')):
self.bows = gensim.corpora.MmCorpus(os.path.join(tmpDir,'corpus.mm'))
if self.use_tfidf:
self.tfidf = gensim.corpora.MmCorpus(os.path.join(tmpDir,'tfidf.mm'))
self.dictionary = Dictionary.load_from_text(os.path.join(tmpDir,'dict.txt'))
self.docs = pickle.load(open(os.path.join(tmpDir,'docs.pkl'),'rb'))
self.dictionary.id2token = {v:k for k,v in self.dictionary.token2id.items()} # because id2token is empty be default, it is a bug.
else:
if stopwords==None:
stopwords = set([l.strip('\n').strip() for l in open(os.path.join(cwd,'data','stopwords.txt'),'r',encoding='utf-8')])
# self.txtLines is the list of string, without any preprocessing.
# self.texts is the list of list of tokens.
print('Tokenizing ...')
tokenizer = Tokenizer if tokenizer==None else tokenizer
self.docs = [tokenizer(txt,stopwords) for txt in tqdm(self.txtLines)]
self.docs = [line for line in self.docs if line!=[]]
# build dictionary
self.dictionary = Dictionary(self.docs)
#self.dictionary.filter_n_most_frequent(remove_n=20)
self.dictionary.filter_extremes(no_below=no_below, no_above=no_above, keep_n=None) # use Dictionary to remove un-relevant tokens
self.dictionary.compactify()
self.dictionary.id2token = {v:k for k,v in self.dictionary.token2id.items()} # because id2token is empty by default, it is a bug.
# convert to BOW representation
self.bows, _docs = [],[]
for doc in self.docs:
_bow = self.dictionary.doc2bow(doc)
if _bow!=[]:
_docs.append(list(doc))
self.bows.append(_bow)
self.docs = _docs
if self.use_tfidf==True:
self.tfidf_model = TfidfModel(self.bows)
self.tfidf = [self.tfidf_model[bow] for bow in self.bows]
# serialize the dictionary
gensim.corpora.MmCorpus.serialize(os.path.join(tmpDir,'corpus.mm'), self.bows)
self.dictionary.save_as_text(os.path.join(tmpDir,'dict.txt'))
pickle.dump(self.docs,open(os.path.join(tmpDir,'docs.pkl'),'wb'))
if self.use_tfidf:
gensim.corpora.MmCorpus.serialize(os.path.join(tmpDir,'tfidf.mm'),self.tfidf)
self.vocabsize = len(self.dictionary)
self.numDocs = len(self.bows)
print(f'Processed {len(self.bows)} documents.')
def __getitem__(self,idx):
bow = torch.zeros(self.vocabsize)
if self.use_tfidf:
item = list(zip(*self.tfidf[idx]))
else:
item = list(zip(*self.bows[idx])) # bow = [[token_id1,token_id2,...],[freq1,freq2,...]]
bow[list(item[0])] = torch.tensor(list(item[1])).float()
txt = self.docs[idx]
return txt,bow
def __len__(self):
return self.numDocs
def collate_fn(self,batch_data):
texts,bows = list(zip(*batch_data))
return texts,torch.stack(bows,dim=0)
def __iter__(self):
for doc in self.docs:
yield doc
def show_dfs_topk(self,topk=20):
ndoc = len(self.docs)
dfs_topk = sorted([(self.dictionary.id2token[k],fq) for k,fq in self.dictionary.dfs.items()],key=lambda x: x[1],reverse=True)[:topk]
for i,(word,freq) in enumerate(dfs_topk):
print(f'{i+1}:{word} --> {freq}/{ndoc} = {(1.0*freq/ndoc):>.13f}')
return dfs_topk
def show_cfs_topk(self,topk=20):
ntokens = sum([v for k,v in self.dictionary.cfs.items()])
cfs_topk = sorted([(self.dictionary.id2token[k],fq) for k,fq in self.dictionary.cfs.items()],key=lambda x: x[1],reverse=True)[:topk]
for i,(word,freq) in enumerate(cfs_topk):
print(f'{i+1}:{word} --> {freq}/{ntokens} = {(1.0*freq/ntokens):>.13f}')
def topk_dfs(self,topk=20):
ndoc = len(self.docs)
dfs_topk = self.show_dfs_topk(topk=topk)
return 1.0*dfs_topk[-1][-1]/ndoc
'''
class DocDataLoader:
def __init__(self,dataset=None,batch_size=128,shuffle=True):
self.dataset = dataset
self.batch_size = batch_size
self.shuffle = shuffle
self.idxes = list(range(len(dataset)))
self.length = len(self.idxes)
def __iter__(self):
return self
def __next__(self):
if self.shuffle==True:
random.shuffle(self.idxes)
for i in range(0,self.length,self.batch_size):
batch_ids = self.idxes[i:i+self.batch_size]
batch_data = self.dataset[batch_ids]
yield batch_data
'''
if __name__ == '__main__':
docSet = DocDataset('zhdd',rebuild=True)
dataloader = DataLoader(docSet,batch_size=64,shuffle=True,num_workers=4,collate_fn=docSet.collate_fn)
print('docSet.docs[10]:',docSet.docs[10])
print(next(iter(dataloader)))
print('The top 20 tokens in document frequency:')
docSet.show_dfs_topk()
print('The top 20 tokens in collections frequency:')
input("Press any key ...")
docSet.show_cfs_topk()
input("Press any key ...")
for doc in docSet:
print(doc)
break
print(docSet.topk_dfs(20))