forked from docker/genai-stack
-
Notifications
You must be signed in to change notification settings - Fork 0
/
loader.py
226 lines (193 loc) · 8.91 KB
/
loader.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
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
import os
import csv
import requests
from dotenv import load_dotenv
from langchain_community.graphs import Neo4jGraph
from langchain_community.vectorstores.chroma import Chroma
from langchain_core.documents.base import Document
import streamlit as st
from streamlit.logger import get_logger
from chains import load_embedding_model
from utils import create_constraints, create_vector_index
from PIL import Image
load_dotenv(".env")
url = os.getenv("NEO4J_URI")
username = os.getenv("NEO4J_USERNAME")
password = os.getenv("NEO4J_PASSWORD")
ollama_base_url = os.getenv("OLLAMA_BASE_URL")
embedding_model_name = os.getenv("EMBEDDING_MODEL")
# Remapping for Langchain Neo4j integration
os.environ["NEO4J_URL"] = url
logger = get_logger(__name__)
so_api_base_url = "https://api.stackexchange.com/2.3/search/advanced"
embeddings, dimension = load_embedding_model(
embedding_model_name, config={"ollama_base_url": ollama_base_url}, logger=logger
)
# if Neo4j is local, you can go to http://localhost:7474/ to browse the database
neo4j_graph = Neo4jGraph(url=url, username=username, password=password)
create_constraints(neo4j_graph)
create_vector_index(neo4j_graph, dimension)
# Initialize ChromaDB
chromadb = Chroma(persist_directory="data_chroma", embedding_function=embeddings)
def load_so_data(tag: str = "neo4j", page: int = 1) -> None:
parameters = (
f"?pagesize=100&page={page}&order=desc&sort=creation&answers=1&tagged={tag}"
"&site=stackoverflow&filter=!*236eb_eL9rai)MOSNZ-6D3Q6ZKb0buI*IVotWaTb"
)
data = requests.get(so_api_base_url + parameters).json()
insert_so_data(data)
def load_high_score_so_data() -> None:
parameters = (
f"?fromdate=1664150400&order=desc&sort=votes&site=stackoverflow&"
"filter=!.DK56VBPooplF.)bWW5iOX32Fh1lcCkw1b_Y6Zkb7YD8.ZMhrR5.FRRsR6Z1uK8*Z5wPaONvyII"
)
data = requests.get(so_api_base_url + parameters).json()
insert_so_data(data)
def insert_so_data(data: dict) -> None:
documents = []
# Prepare directory for storing CSV
os.makedirs('data_openai', exist_ok=True)
with open('data_openai/data.csv', mode='a', newline='', encoding='utf-8') as file:
csv_writer = csv.writer(file)
# Write headers if the file is new/empty
if file.tell() == 0:
csv_writer.writerow([
"Question Title", "Question Body", "Answer Body", "Score", "Link"
#"Is Accepted", "Reputation", "Favorite Count", "Q Creation Date",
#"A Creation Date", "Answer ID", "Question ID", "Tags"
])
# Calculate embedding values for questions and answers
for q in data["items"]:
# Neo4j
question_text = q["title"] + "\n" + q["body_markdown"]
question_embedding = embeddings.embed_query(question_text)
q["embedding"] = question_embedding
for a in q["answers"]:
# Neo4j
answer_text = question_text + "\n" + a["body_markdown"]
answer_embedding = embeddings.embed_query(answer_text)
a["embedding"] = answer_embedding
# ChromaDB Q&A pairs
QA_document = Document(
page_content = "Question: " + q.get("title", None) + '\n' + \
q.get("body_markdown", None) + '\n' + \
"Answer:" + '\n' + \
a.get("body_markdown", None) + '\n' + \
"Score:" + '\n' + \
str(a.get("score", 0)) + '\n' + \
"Link:" + '\n' + \
q.get("link", None),
metadata={
"source": "stackoverflow",
"title": q.get("title", None),
"score": a.get("score", 0),
"link": q.get("link", None),
"is_accepted": a.get("is_accepted", False),
"reputation": a.get("owner", {}).get("reputation", 0),
"favorite_count": q.get("favorite_count", 0),
"q_creation_date": q.get("creation_date", None),
"a_creation_date": a.get("creation_date", None),
"answer_id": a.get("answer_id", None),
"question_id": q.get("question_id", None),
"tags": ','.join(q.get("tags", [])),
}
)
documents.append(QA_document)
# Writing data to CSV
csv_writer.writerow([
q.get("title", ""),
q.get("body_markdown", ""),
a.get("body_markdown", ""),
a.get("score", 0),
q.get("link", ""),
#a.get("is_accepted", False),
#a.get("owner", {}).get("reputation", 0),
#q.get("favorite_count", 0),
#q.get("creation_date", ""),
#a.get("creation_date", ""),
#a.get("answer_id", ""),
#q.get("question_id", ""),
#','.join(q.get("tags", []))
])
# Write the page_content to a file for OpenAI RAG API
os.makedirs('data_openai', exist_ok=True)
with open('data_openai/data.txt', mode='a') as f:
f.write(QA_document.page_content + '\n\n')
# Cypher, the query language of Neo4j, is used to import the data
# https://neo4j.com/docs/getting-started/cypher-intro/
# https://neo4j.com/docs/cypher-cheat-sheet/5/auradb-enterprise/
import_query = """
UNWIND $data AS q
MERGE (question:Question {id:q.question_id})
ON CREATE SET question.title = q.title, question.link = q.link, question.score = q.score,
question.favorite_count = q.favorite_count, question.creation_date = datetime({epochSeconds: q.creation_date}),
question.body = q.body_markdown, question.embedding = q.embedding
FOREACH (tagName IN q.tags |
MERGE (tag:Tag {name:tagName})
MERGE (question)-[:TAGGED]->(tag)
)
FOREACH (a IN q.answers |
MERGE (question)<-[:ANSWERS]-(answer:Answer {id:a.answer_id})
SET answer.is_accepted = a.is_accepted,
answer.score = a.score,
answer.creation_date = datetime({epochSeconds:a.creation_date}),
answer.body = a.body_markdown,
answer.embedding = a.embedding
MERGE (answerer:User {id:coalesce(a.owner.user_id, "deleted")})
ON CREATE SET answerer.display_name = a.owner.display_name,
answerer.reputation = a.owner.reputation
MERGE (answer)<-[:PROVIDED]-(answerer)
)
WITH * WHERE NOT q.owner.user_id IS NULL
MERGE (owner:User {id:q.owner.user_id})
ON CREATE SET owner.display_name = q.owner.display_name,
owner.reputation = q.owner.reputation
MERGE (owner)-[:ASKED]->(question)
"""
neo4j_graph.query(import_query, {"data": data["items"]})
# Insert data into ChromaDB
chromadb.add_documents(documents=documents)
chromadb.persist()
# Streamlit
def get_tag() -> str:
input_text = st.text_input(
"Which tag questions do you want to import?", value="neo4j"
)
return input_text
def get_pages():
col1, col2 = st.columns(2)
with col1:
num_pages = st.number_input(
"Number of pages (100 questions per page)", step=1, min_value=1
)
with col2:
start_page = st.number_input("Start page", step=1, min_value=1)
st.caption("Only questions with answers will be imported.")
return (int(num_pages), int(start_page))
def render_page():
datamodel_image = Image.open("./images/datamodel.png")
st.header("StackOverflow Loader")
st.subheader("Choose StackOverflow tags to load into Neo4j")
st.caption("Go to http://localhost:7474/ to explore the graph.")
user_input = get_tag()
num_pages, start_page = get_pages()
if st.button("Import", type="primary"):
with st.spinner("Loading... This might take a minute or two."):
try:
for page in range(1, num_pages + 1):
load_so_data(user_input, start_page + (page - 1))
st.success("Import successful", icon="✅")
st.caption("Data model")
st.image(datamodel_image)
st.caption("Go to http://localhost:7474/ to interact with the database")
except Exception as e:
st.error(f"Error: {e}", icon="🚨")
with st.expander("Highly ranked questions rather than tags?"):
if st.button("Import highly ranked questions"):
with st.spinner("Loading... This might take a minute or two."):
try:
load_high_score_so_data()
st.success("Import successful", icon="✅")
except Exception as e:
st.error(f"Error: {e}", icon="🚨")
render_page()