-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathindexing_livedoor.py
209 lines (173 loc) · 7.64 KB
/
indexing_livedoor.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
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
# chromadb
import chromadb
from chromadb.config import Settings
from chromadb.utils import embedding_functions
# langchain
from langchain.document_loaders import DirectoryLoader
from langchain.document_loaders import TextLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
# time
import time
# sys
import sys
# pprint
import pprint
# url detect
import re
# tqdm
import tqdm
from tqdm import tqdm
import numpy as np
#directory = './livedoorニュースコーパス/ldcc-20140209/text/small/'
#directory = './livedoorニュースコーパス/ldcc-20140209/text/movie-enter/'
directory = './livedoorニュースコーパス/ldcc-20140209/text/it-life-hack/'
#loader = DirectoryLoader(directory, glob="movie-enter*.txt",show_progress=True, encoding='utf8')
#loader = TextLoader(directory+"movie-enter-5840081.txt", encoding='utf8')
#loader = DirectoryLoader(directory, glob="movie-enter*.txt",show_progress=True, loader_cls=TextLoader, encoding='utf8')
#loader = DirectoryLoader(directory, glob="movie-enter*.txt", loader_cls=TextLoader, encoding='utf8')
# reffer to https://python.langchain.com/docs/modules/data_connection/document_loaders/file_directory
# C.Auto detect encodings
text_loader_kwargs={'autodetect_encoding': True}
#loader = DirectoryLoader(directory, glob="movie-enter*.txt", show_progress=True, loader_cls=TextLoader, loader_kwargs=text_loader_kwargs)
loader = DirectoryLoader(directory, glob="it-life-hack*.txt", show_progress=True, loader_cls=TextLoader, loader_kwargs=text_loader_kwargs)
start = time.time()
documents = loader.load()
end = time.time()
time_diff = end - start
num_documents = len(documents)
spendding_time = round(time_diff,2)
print('documents:{0} documents spendding {1} seconds.'.format(num_documents,str(round(time_diff,2))))
# scheme of documents
# document[0]
# document[0].page_content
# document[0].metadata
# ...
# document[10]
# document[10].page_content
# document[10].metadata
# for document in documents:
# print("path:{0}".format(document.metadata['source']))
# pprint.pprint(documents[0: 2],indent=2,width=40)
# pprint.pprint(documents[0].page_content,indent=2,width=40)
# pprint.pprint(documents[0].metadata,indent=2,width=40)
# Number of documents processed at once
num_proc_documents = 20
cur_document_num = 0
bar = tqdm(total = num_documents)
bar.set_description('Progress rate')
start = time.time()
for i in range(0,num_documents,num_proc_documents):
# tqdm.write("i:{0}".format(i))
# tqdm.write("i+num_proc_documents:{0}".format(i+(num_proc_documents)))
# tqdm.write("num_documents:{0}".format(num_documents))
# if(num_documents % num_proc_documents):
# num_documents
proc_documents = documents[cur_document_num:(cur_document_num+num_proc_documents):1]
# tqdm.write("proc_documents[{0}:{1}:1]".format(cur_document_num,cur_document_num+num_proc_documents))
# tqdm.write("num proc_documents:{0}".format(len(proc_documents)))
#pprint.pprint(proc_documents,indent=2,width=40)
# for doc in proc_documents:
# tqdm.write("path:{0}".format(doc.metadata['source']))
#pprint.pprint(proc_documents,indent=2,width=40)
try:
#client = chromadb.HttpClient(
# host='localhost',
# port=80)
# With authentifizations
client = chromadb.HttpClient(
host='localhost',
port=80,
settings=Settings(chroma_client_auth_provider='chromadb.auth.token.TokenAuthClientProvider',
chroma_client_auth_credentials='test-token'))
except Exception as e:
print('Vector database Connection error occurs with following message.')
print('Error Message:{0}'.format(str(e)))
sys.exit(-1)
# defined sentence transformer LLM
sentence_transformer_ef = embedding_functions.SentenceTransformerEmbeddingFunction(model_name="intfloat/multilingual-e5-large")
# get or create collection
collection = client.get_or_create_collection("livedoor",embedding_function=sentence_transformer_ef)
items = collection.get()
#pprint.pprint(items,indent=2,width=40)
if(len(items['ids']) == 0):
last_ids = 0
else:
last_ids = int(items['ids'][-1])
#print('last ids:{0}'.format(last_ids))
# split chunk each num_proc_documents
chunk_size=512
chunk_overlap=20
text_splitter = RecursiveCharacterTextSplitter(chunk_size=chunk_size, chunk_overlap=chunk_overlap)
chunks = text_splitter.split_documents(proc_documents)
tqdm.write("{0} chunks in {1} documents".format(len(chunks),len(proc_documents)))
# scheme of Chunk
# chunks[0]
# chunks[0].page_content
# "http://news.livedoor.com/article/detail/5840081/\n\n2011\n\n09\n\n08T10:00:00+0900\n\nインタビュー：宮崎あおい..."
# chunks[0].metadata
# {'source':'livedoorニュースコーパス\\ldcc-20140209\\text\\small\\movie-enter-5840081.txt'}
# ...
# chunks[10]
# chunks[10].page_content
# chunks[10].metadata
# Initialization Scheme of Vector Database
vect_documents = []
vect_metadatas = []
vect_ids = []
# Restructure chunks
cur_chunk_num = 0
# defined current ids number
cur_ids = last_ids + 1
for chunk in chunks:
# strip_docs.append(chunk.page_content)
splitline_chunk_page_content = chunk.page_content.splitlines()
check_url = re.findall('^https?://[\w/:%#\$&\?\(\)~\.=\+\-]+', splitline_chunk_page_content[0])
if(check_url):
# first chunk of document
# print("found url:{0}".format(check_url[0]))
# tqdm.write("document_url :{0}".format(check_url[0]))
document_url = check_url[0]
document_date = splitline_chunk_page_content[1]
document_title = splitline_chunk_page_content[2]
target_page_content = splitline_chunk_page_content[3]
#cur_document = chunk.page_content[index_document+1]
#cur_document = chunk.page_content
index_document = chunk.page_content.find(target_page_content)
#print(f'target_page_content:{target_page_content}')
#print(f'index_document :{index_document}')
cur_document = chunk.page_content[index_document:]
cur_chunk_num = 0
#bar.update(cur_document_num/num_documents)
cur_document_num+=1
# tqdm.write("cur_document_num:{0}".format(cur_document_num))
bar.update(1)
else:
# Second and subsequent chunks
cur_document = chunk.page_content
# print(f'cur_chunk_num:{cur_chunk_num}')
# tqdm.write(f'cur_chunk_num:{cur_chunk_num}')
# print(f'cur_document:{cur_document}')
cur_document = cur_document.replace('\u200b', '')
cur_document = cur_document.replace('\u3000', '')
vect_documents.append(cur_document)
dict_metadatas = {}
dict_metadatas["url"] = document_url
dict_metadatas["date"] = document_date
document_title = document_title.replace('\u200b', '')
document_title = document_title.replace('\u3000', '')
dict_metadatas["title"] = document_title
dict_metadatas["chunk"] = cur_chunk_num
vect_metadatas.append(dict_metadatas)
vect_ids.append(str(cur_ids))
cur_ids+=1
collection.add(
ids=vect_ids,
metadatas=vect_metadatas,
documents=vect_documents
)
cur_chunk_num = cur_chunk_num + 1
#bar.update(cur_document_num/num_documents)
#bar.update(cur_document_num)
end = time.time()
time_diff = end - start
print('documents:{0} documents spendding {1} seconds.'.format(num_documents,str(round(time_diff,2))))