Upload 3 files
Browse files- .gitattributes +1 -0
- The Elements of Statisitcal Learning.pdf +3 -0
- app.py +213 -0
- requirements.txt +10 -0
.gitattributes
CHANGED
@@ -32,3 +32,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
32 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
33 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
34 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
|
|
|
32 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
33 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
34 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
35 |
+
The[[:space:]]Elements[[:space:]]of[[:space:]]Statisitcal[[:space:]]Learning.pdf filter=lfs diff=lfs merge=lfs -text
|
The Elements of Statisitcal Learning.pdf
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:1da55e44889d3c45c0e0068b8b953ba71a95a85784485b0d14f52f34bb7f9eca
|
3 |
+
size 13303613
|
app.py
ADDED
@@ -0,0 +1,213 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import re
|
2 |
+
import numpy as np
|
3 |
+
import tensorflow_hub as hub
|
4 |
+
import openai
|
5 |
+
import os
|
6 |
+
import tensorflow_text
|
7 |
+
from sklearn.neighbors import NearestNeighbors
|
8 |
+
import gradio as gr
|
9 |
+
import requests
|
10 |
+
import json
|
11 |
+
import fitz
|
12 |
+
|
13 |
+
#这里填写调用openai需要的密钥
|
14 |
+
openai.api_key = '9481961416fa4c8e883047c5679cf971'
|
15 |
+
openai.api_base = 'https://demopro-oai-we2.openai.azure.com/'
|
16 |
+
openai.api_type = 'azure'
|
17 |
+
openai.api_version = '2022-12-01'
|
18 |
+
|
19 |
+
#将嵌套的列表展平
|
20 |
+
def flatten(_2d_list):
|
21 |
+
flat_list = []
|
22 |
+
for element in _2d_list:
|
23 |
+
if type(element) is list:
|
24 |
+
for item in element:
|
25 |
+
flat_list.append(item)
|
26 |
+
else:
|
27 |
+
flat_list.append(element)
|
28 |
+
return flat_list
|
29 |
+
|
30 |
+
|
31 |
+
def preprocess(text):
|
32 |
+
text = text.replace('\n', ' ')
|
33 |
+
text = re.sub('\s+', ' ', text)
|
34 |
+
return text
|
35 |
+
|
36 |
+
#将pdf文档按段落分
|
37 |
+
# def pdf_to_text(path):
|
38 |
+
# doc = pdfplumber.open(path)
|
39 |
+
# pages = doc.pages
|
40 |
+
# text_list=[]
|
41 |
+
# for page,d in enumerate(pages):
|
42 |
+
# d=d.extract_text()
|
43 |
+
# d=preprocess(d)
|
44 |
+
# text_list.append(d)
|
45 |
+
# doc.close()
|
46 |
+
|
47 |
+
# return text_list
|
48 |
+
|
49 |
+
|
50 |
+
|
51 |
+
def pdf_to_text(path, start_page=1, end_page=None):
|
52 |
+
doc = fitz.open(path)
|
53 |
+
total_pages = doc.page_count
|
54 |
+
|
55 |
+
if end_page is None:
|
56 |
+
end_page = total_pages
|
57 |
+
|
58 |
+
text_list = []
|
59 |
+
|
60 |
+
for i in range(start_page - 1, end_page):
|
61 |
+
text = doc.load_page(i).get_text("text")
|
62 |
+
text = preprocess(text)
|
63 |
+
text_list.append(text)
|
64 |
+
|
65 |
+
doc.close()
|
66 |
+
return text_list
|
67 |
+
|
68 |
+
|
69 |
+
def text_to_chunks(texts, word_length=150, start_page=1):
|
70 |
+
text_toks = [t.split(' ') for t in texts]
|
71 |
+
page_nums = []
|
72 |
+
chunks = []
|
73 |
+
|
74 |
+
for idx, words in enumerate(text_toks):
|
75 |
+
for i in range(0, len(words), word_length):
|
76 |
+
chunk = words[i : i + word_length]
|
77 |
+
if (
|
78 |
+
(i + word_length) > len(words)
|
79 |
+
and (len(chunk) < word_length)
|
80 |
+
and (len(text_toks) != (idx + 1))
|
81 |
+
):
|
82 |
+
text_toks[idx + 1] = chunk + text_toks[idx + 1]
|
83 |
+
continue
|
84 |
+
chunk = ' '.join(chunk).strip()
|
85 |
+
chunk = f'[Page no. {idx+start_page}]' + ' ' + '"' + chunk + '"'
|
86 |
+
chunks.append(chunk)
|
87 |
+
return chunks
|
88 |
+
|
89 |
+
history=pdf_to_text('The Elements of Statisitcal Learning.pdf',start_page=20)
|
90 |
+
history=text_to_chunks(history,start_page=1)
|
91 |
+
|
92 |
+
|
93 |
+
def encoder(text):
|
94 |
+
embed=openai.Embedding.create(input=text, engine="text-embedding-ada-002")
|
95 |
+
return embed.get('data')[0].get('embedding')
|
96 |
+
|
97 |
+
|
98 |
+
#定义语义搜索类
|
99 |
+
class SemanticSearch:
|
100 |
+
|
101 |
+
def __init__(self):
|
102 |
+
#类初始化,使用google公司的多语言语句编码,第一次运行时需要十几分钟的时间下载
|
103 |
+
self.use =hub.load('https://tfhub.dev/google/universal-sentence-encoder-multilingual/3')
|
104 |
+
self.fitted = False
|
105 |
+
|
106 |
+
def get_text_embedding(self, texts, batch=1000):
|
107 |
+
embeddings = []
|
108 |
+
for i in range(0, len(texts), batch):
|
109 |
+
text_batch = texts[i : (i + batch)]
|
110 |
+
emb_batch = self.use(text_batch)
|
111 |
+
embeddings.append(emb_batch)
|
112 |
+
embeddings = np.vstack(embeddings)
|
113 |
+
return embeddings
|
114 |
+
|
115 |
+
|
116 |
+
|
117 |
+
#K近邻算法,找到与问题最相似的 k 个段落,这里的 k 即n_neighbors=10
|
118 |
+
def fit(self, data, batch=1000, n_neighbors=5):
|
119 |
+
self.data = data
|
120 |
+
self.embeddings = self.get_text_embedding(data, batch=batch)
|
121 |
+
n_neighbors = min(n_neighbors, len(self.embeddings))
|
122 |
+
self.nn = NearestNeighbors(n_neighbors=n_neighbors)
|
123 |
+
self.nn.fit(self.embeddings)
|
124 |
+
self.fitted = True
|
125 |
+
|
126 |
+
#定义了该方法后,实例就可以被当作函数调用,text参数即用户提出的问题,inp_emb为其转化成的向量
|
127 |
+
def __call__(self, text, return_data=True):
|
128 |
+
inp_emb = self.use([text])
|
129 |
+
|
130 |
+
|
131 |
+
neighbors = self.nn.kneighbors(inp_emb, return_distance=False)[0]
|
132 |
+
|
133 |
+
if return_data:
|
134 |
+
return [self.data[i] for i in neighbors]
|
135 |
+
else:
|
136 |
+
return neighbors
|
137 |
+
|
138 |
+
|
139 |
+
#openai的api接口,engine参数为我们选择的大语言模型,prompt即提示词
|
140 |
+
def generate_text(prompt, engine="text-davinci-003"):
|
141 |
+
completions = openai.Completion.create(
|
142 |
+
engine=engine,
|
143 |
+
prompt=prompt,
|
144 |
+
max_tokens=512,
|
145 |
+
n=1,
|
146 |
+
stop=None,
|
147 |
+
temperature=0.7,
|
148 |
+
)
|
149 |
+
message = completions.choices[0].text
|
150 |
+
return message
|
151 |
+
|
152 |
+
|
153 |
+
def generate_answer(question):
|
154 |
+
#匹配与问题最相近的n个段落,前面定义了n=10
|
155 |
+
topn_chunks = recommender(question)
|
156 |
+
prompt = ""
|
157 |
+
prompt += 'search results:\n\n'
|
158 |
+
|
159 |
+
#把匹配到的段落加进提示词
|
160 |
+
for c in topn_chunks:
|
161 |
+
prompt += c + '\n\n'
|
162 |
+
|
163 |
+
#提示词
|
164 |
+
prompt += '''
|
165 |
+
Instructions: 如果搜索结果中找不到相关信息,只需要回答'未在该文档中找到相关信息'。
|
166 |
+
如果找到了相关信息,请使用中文回答,回答尽量精确简洁。并在句子的末尾使用[七年级上册/七年级下册页码]符号引用每个参考文献(每个结果的开头都有这个编号)
|
167 |
+
如果不确定答案是否正确,就仅给出相似段落的来源,不要回复错误的答案。
|
168 |
+
\n\nQuery: {question}\nAnswer:
|
169 |
+
'''
|
170 |
+
|
171 |
+
prompt += f"Query: {question}\nAnswer:"
|
172 |
+
answer = generate_text(prompt,"text-davinci-003")
|
173 |
+
return answer
|
174 |
+
|
175 |
+
|
176 |
+
recommender = SemanticSearch()
|
177 |
+
recommender.fit(history)
|
178 |
+
|
179 |
+
|
180 |
+
#以下为web客户端搭建,运行后产生客户端界面
|
181 |
+
def ask_api(question):
|
182 |
+
|
183 |
+
if question.strip() == '':
|
184 |
+
return '[ERROR]: 未输入问题'
|
185 |
+
|
186 |
+
return generate_answer(question)
|
187 |
+
|
188 |
+
title = 'Chat With Statistical Learning'
|
189 |
+
description = """ 该机器人将以Trevor Hastie等人所著的The Elements of Statistical Learning Data Mining, Inference, and Prediction
|
190 |
+
(即我们上课所用的课本)为主题回答你的问题,如果所问问题与书的内容无关,将会返回"未在该文档中找到相关信息"
|
191 |
+
"""
|
192 |
+
|
193 |
+
with gr.Blocks() as demo:
|
194 |
+
gr.Markdown(f'<center><h1>{title}</h1></center>')
|
195 |
+
gr.Markdown(description)
|
196 |
+
|
197 |
+
with gr.Row():
|
198 |
+
with gr.Group():
|
199 |
+
question = gr.Textbox(label='请输入你的问题')
|
200 |
+
btn = gr.Button(value='提交')
|
201 |
+
btn.style(full_width=True)
|
202 |
+
|
203 |
+
with gr.Group():
|
204 |
+
answer = gr.Textbox(label='回答:')
|
205 |
+
|
206 |
+
btn.click(
|
207 |
+
ask_api,
|
208 |
+
inputs=[question],
|
209 |
+
outputs=[answer]
|
210 |
+
)
|
211 |
+
|
212 |
+
#参数share=True会产生一个公开网页,别人可以通过访问该网页使用你的模型,前提是你需要正在运行这段代码(将自己的电脑当作服务器)
|
213 |
+
demo.launch()
|
requirements.txt
ADDED
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
fitz
|
2 |
+
gradio
|
3 |
+
tensorflow
|
4 |
+
tensorflow_hub
|
5 |
+
tensorflow_text
|
6 |
+
pdfplumber
|
7 |
+
openai
|
8 |
+
numpy
|
9 |
+
requests
|
10 |
+
scikit-learn
|