Debate / src /agents /Component /ToolComponent.py
callanwu's picture
update
2441227
raw
history blame
29.8 kB
from abc import abstractmethod
import uuid
from text2vec import semantic_search
from utils import (
get_relevant_history,
load_knowledge_base_qa,
load_knowledge_base_UnstructuredFile,
get_embedding,
extract,
)
import json
from typing import Dict, List
import os
from googleapiclient.discovery import build
import requests
from selenium import webdriver
from selenium.webdriver.common.by import By
from selenium.webdriver.support.ui import WebDriverWait
from selenium.webdriver.support import expected_conditions as EC
from bs4 import BeautifulSoup
import base64
import re
from datetime import datetime, timedelta
from typing import Tuple, List, Any, Dict
from email.mime.text import MIMEText
from email.mime.multipart import MIMEMultipart
from google.auth.transport.requests import Request
from google.oauth2.credentials import Credentials
from google_auth_oauthlib.flow import InstalledAppFlow
from googleapiclient.discovery import build
from googleapiclient.errors import HttpError
from tqdm import tqdm
class ToolComponent:
def __init__(self):
pass
@abstractmethod
def func(self):
pass
class KnowledgeBaseComponent(ToolComponent):
"""
Inject knowledge base
top_k : Top_k with the highest matching degree
type : "QA" or others
knowledge_base(json_path) : knowledge_base_path
"""
def __init__(self, top_k, type, knowledge_base):
super().__init__()
self.top_k = top_k
self.type = type
self.knowledge_base = knowledge_base
if self.type == "QA":
(
self.kb_embeddings,
self.kb_questions,
self.kb_answers,
self.kb_chunks,
) = load_knowledge_base_qa(self.knowledge_base)
else:
self.kb_embeddings, self.kb_chunks = load_knowledge_base_UnstructuredFile(
self.knowledge_base
)
def func(self, agent):
query = (
agent.long_term_memory[-1]["content"]
if len(agent.long_term_memory) > 0
else ""
)
knowledge = ""
query = extract(query, "query")
query_embedding = get_embedding(query)
hits = semantic_search(query_embedding, self.kb_embeddings, top_k=50)
hits = hits[0]
temp = []
if self.type == "QA":
for hit in hits:
matching_idx = hit["corpus_id"]
if self.kb_chunks[matching_idx] in temp:
pass
else:
knowledge = (
knowledge
+ f"question:{self.kb_questions[matching_idx]},answer:{self.kb_answers[matching_idx]}\n\n"
)
temp.append(self.kb_answers[matching_idx])
if len(temp) == 1:
break
print(hits[0]["score"])
score = hits[0]["score"]
if score < 0.5:
return {"prompt": "No matching knowledge base"}
else:
return {"prompt": "The relevant content is: " + knowledge + "\n"}
else:
for hit in hits:
matching_idx = hit["corpus_id"]
if self.kb_chunks[matching_idx] in temp:
pass
else:
knowledge = knowledge + f"{self.kb_answers[matching_idx]}\n\n"
temp.append(self.kb_answers[matching_idx])
if len(temp) == self.top_k:
break
print(hits[0]["score"])
score = hits[0]["score"]
if score < 0.5:
return {"prompt": "No matching knowledge base"}
else:
print(knowledge)
return {"prompt": "The relevant content is: " + knowledge + "\n"}
class StaticComponent(ToolComponent):
"Return static response"
def __init__(self, output):
super().__init__()
self.output = output
def func(self, agent):
outputdict = {"response": self.output}
return outputdict
class ExtractComponent(ToolComponent):
"""
Extract keywords based on the current scene and store them in the environment
extract_words(list) : Keywords to be extracted
system_prompt & last_prompt : Prompt to extract keywords
"""
def __init__(
self,
extract_words,
system_prompt,
last_prompt=None,
):
super().__init__()
self.extract_words = extract_words
self.system_prompt = system_prompt
self.default_prompt = (
"Please strictly adhere to the following format for outputting:\n"
)
for extract_word in extract_words:
self.default_prompt += (
f"<{extract_word}> the content you need to extract </{extract_word}>"
)
self.last_prompt = last_prompt if last_prompt else self.default_prompt
def func(self, agent):
response = agent.LLM.get_response(
agent.long_term_memory,
self.system_prompt,
self.last_prompt,
stream=False,
)
for extract_word in self.extract_words:
key = extract(response, extract_word)
key = key if key else response
agent.environment.shared_memory[extract_word] = key
return {}
"""Search sources: chatgpt/search engines/specific search sources/can even be multimodal (if it comes to clothing)"""
class WebSearchComponent(ToolComponent):
"""search engines"""
__ENGINE_NAME__: List = ["google", "bing"]
def __init__(self, engine_name: str, api: Dict):
"""
:param engine_name: The name of the search engine used
:param api: Pass in a dictionary, such as {"bing":"key1", "google":"key2", ...}, of course each value can also be a list, or more complicated
"""
super(WebSearchComponent, self).__init__()
"""Determine whether the key and engine_name of the api are legal"""
assert engine_name in WebSearchComponent.__ENGINE_NAME__
for api_name in api:
assert api_name in WebSearchComponent.__ENGINE_NAME__
self.api = api
self.engine_name = engine_name
self.search: Dict = {"bing": self._bing_search, "google": self._google_search}
def _bing_search(self, query: str, **kwargs):
"""Initialize search hyperparameters"""
subscription_key = self.api["bing"]
search_url = "https://api.bing.microsoft.com/v7.0/search"
headers = {"Ocp-Apim-Subscription-Key": subscription_key}
params = {
"q": query,
"textDecorations": True,
"textFormat": "HTML",
"count": 10,
}
"""start searching"""
response = requests.get(search_url, headers=headers, params=params)
response.raise_for_status()
results = response.json()["webPages"]["value"]
"""execute"""
metadata_results = []
for result in results:
metadata_result = {
"snippet": result["snippet"],
"title": result["name"],
"link": result["url"],
}
metadata_results.append(metadata_result)
return {"meta data": metadata_results}
def _google_search(self, query: str, **kwargs):
"""Initialize search hyperparameters"""
api_key = self.api[self.engine_name]["api_key"]
cse_id = self.api[self.engine_name]["cse_id"]
service = build("customsearch", "v1", developerKey=api_key)
"""start searching"""
results = (
service.cse().list(q=query, cx=cse_id, num=10, **kwargs).execute()["items"]
)
"""execute"""
metadata_results = []
for result in results:
metadata_result = {
"snippet": result["snippet"],
"title": result["title"],
"link": result["link"],
}
metadata_results.append(metadata_result)
return {"meta data": metadata_results}
def func(self, agent, **kwargs) -> Dict:
query = (
agent.long_term_memory[-1]["content"]
if len(agent.long_term_memory) > 0
else " "
)
response = agent.LLM.get_response(
None,
system_prompt=f"Please analyze the provided conversation and identify keywords that can be used for a search engine query. Format the output as <keywords>extracted keywords</keywords>:\nConversation:\n{query}",
stream=False,
)
response = extract(response, "keywords")
query = response if response else query
search_results = self.search[self.engine_name](query=query, **kwargs)
information = ""
for i in search_results["meta data"][:5]:
information += i["snippet"]
return {
"prompt": "You can refer to the following information to reply:\n"
+ information
}
def convert_search_engine_to(self, engine_name):
assert engine_name in WebSearchComponent.__ENGINE_NAME__
self.engine_name = engine_name
class WebCrawlComponent(ToolComponent):
"""Open a single web page for crawling"""
def __init__(self):
super(WebCrawlComponent, self).__init__()
def func(self, agent_dict) -> Dict:
url = agent_dict["url"]
print(f"crawling {url} ......")
content = ""
"""Crawling content from url may need to be carried out according to different websites, such as wiki, baidu, zhihu, etc."""
driver = webdriver.Chrome()
try:
"""open url"""
driver.get(url)
"""wait 20 second"""
wait = WebDriverWait(driver, 20)
wait.until(EC.presence_of_element_located((By.TAG_NAME, "body")))
"""crawl code"""
page_source = driver.page_source
"""parse"""
soup = BeautifulSoup(page_source, "html.parser")
"""concatenate"""
for paragraph in soup.find_all("p"):
content = f"{content}\n{paragraph.get_text()}"
except Exception as e:
print("Error:", e)
finally:
"""quit"""
driver.quit()
return {"content": content.strip()}
class MailComponent(ToolComponent):
__VALID_ACTION__ = ["read", "send"]
def __init__(
self, cfg_file: str, default_action: str = "read", name: str = "e-mail"
):
"""'../config/google_mail.json'"""
super(MailComponent, self).__init__(name)
self.name = name
assert (
default_action.lower() in self.__VALID_ACTION__
), f"Action `{default_action}` is not allowed! The valid action is in `{self.__VALID_ACTION__}`"
self.action = default_action.lower()
self.credential = self._login(cfg_file)
def _login(self, cfg_file: str):
SCOPES = [
"https://www.googleapis.com/auth/gmail.readonly",
"https://www.googleapis.com/auth/gmail.send",
]
creds = None
if os.path.exists("token.json"):
print("Login Successfully!")
creds = Credentials.from_authorized_user_file("token.json", SCOPES)
if not creds or not creds.valid:
print("Please authorize in an open browser.")
if creds and creds.expired and creds.refresh_token:
creds.refresh(Request())
else:
flow = InstalledAppFlow.from_client_secrets_file(cfg_file, SCOPES)
creds = flow.run_local_server(port=0)
# Save the credentials for the next run
with open("token.json", "w") as token:
token.write(creds.to_json())
return creds
def _read(self, mail_dict: dict):
credential = self.credential
state = mail_dict["state"] if "state" in mail_dict else None
time_between = (
mail_dict["time_between"] if "time_between" in mail_dict else None
)
sender_mail = mail_dict["sender_mail"] if "sender_mail" in mail_dict else None
only_both = mail_dict["only_both"] if "only_both" in mail_dict else False
order_by_time = (
mail_dict["order_by_time"] if "order_by_time" in mail_dict else "descend"
)
include_word = (
mail_dict["include_word"] if "include_word" in mail_dict else None
)
exclude_word = (
mail_dict["exclude_word"] if "exclude_word" in mail_dict else None
)
MAX_SEARCH_CNT = (
mail_dict["MAX_SEARCH_CNT"] if "MAX_SEARCH_CNT" in mail_dict else 50
)
number = mail_dict["number"] if "number" in mail_dict else 10
if state is None:
state = "all"
if time_between is not None:
assert isinstance(time_between, tuple)
assert len(time_between) == 2
assert state in ["all", "unread", "read", "sent"]
if only_both:
assert sender_mail is not None
if sender_mail is not None:
assert isinstance(sender_mail, str)
assert credential
assert order_by_time in ["descend", "ascend"]
def generate_query():
query = ""
if state in ["unread", "read"]:
query = f"is:{state}"
if state in ["sent"]:
query = f"in:{state}"
if only_both:
query = f"{query} from:{sender_mail} OR to:{sender_mail}"
if sender_mail is not None and not only_both:
query = f"{query} from:({sender_mail})"
if include_word is not None:
query = f"{query} {include_word}"
if exclude_word is not None:
query = f"{query} -{exclude_word}"
if time_between is not None:
TIME_FORMAT = "%Y/%m/%d"
t1, t2 = time_between
if t1 == "now":
t1 = datetime.now().strftime(TIME_FORMAT)
if t2 == "now":
t2 = datetime.now().strftime(TIME_FORMAT)
if isinstance(t1, str) and isinstance(t2, str):
t1 = datetime.strptime(t1, TIME_FORMAT)
t2 = datetime.strptime(t2, TIME_FORMAT)
elif isinstance(t1, str) and isinstance(t2, int):
t1 = datetime.strptime(t1, TIME_FORMAT)
t2 = t1 + timedelta(days=t2)
elif isinstance(t1, int) and isinstance(t2, str):
t2 = datetime.strptime(t2, TIME_FORMAT)
t1 = t2 + timedelta(days=t1)
else:
assert False, "invalid time"
if t1 > t2:
t1, t2 = t2, t1
query = f"{query} after:{t1.strftime(TIME_FORMAT)} before:{t2.strftime(TIME_FORMAT)}"
return query.strip()
def sort_by_time(data: List[Dict]):
if order_by_time == "descend":
reverse = True
else:
reverse = False
sorted_data = sorted(
data,
key=lambda x: datetime.strptime(x["time"], "%Y-%m-%d %H:%M:%S"),
reverse=reverse,
)
return sorted_data
try:
service = build("gmail", "v1", credentials=credential)
results = (
service.users()
.messages()
.list(userId="me", labelIds=["INBOX"], q=generate_query())
.execute()
)
messages = results.get("messages", [])
email_data = list()
if not messages:
print("No eligible emails.")
return None
else:
pbar = tqdm(total=min(MAX_SEARCH_CNT, len(messages)))
for cnt, message in enumerate(messages):
pbar.update(1)
if cnt >= MAX_SEARCH_CNT:
break
msg = (
service.users()
.messages()
.get(
userId="me",
id=message["id"],
format="full",
metadataHeaders=None,
)
.execute()
)
subject = ""
for header in msg["payload"]["headers"]:
if header["name"] == "Subject":
subject = header["value"]
break
sender = ""
for header in msg["payload"]["headers"]:
if header["name"] == "From":
sender = re.findall(
r"\b[\w\.-]+@[\w\.-]+\.\w+\b", header["value"]
)[0]
break
body = ""
if "parts" in msg["payload"]:
for part in msg["payload"]["parts"]:
if part["mimeType"] == "text/plain":
data = part["body"]["data"]
body = base64.urlsafe_b64decode(data).decode("utf-8")
break
email_info = {
"sender": sender,
"time": datetime.fromtimestamp(
int(msg["internalDate"]) / 1000
).strftime("%Y-%m-%d %H:%M:%S"),
"subject": subject,
"body": body,
}
email_data.append(email_info)
pbar.close()
email_data = sort_by_time(email_data)[0:number]
return {"results": email_data}
except Exception as e:
print(e)
return None
def _send(self, mail_dict: dict):
recipient_mail = mail_dict["recipient_mail"]
subject = mail_dict["subject"]
body = mail_dict["body"]
credential = self.credential
service = build("gmail", "v1", credentials=credential)
message = MIMEMultipart()
message["to"] = recipient_mail
message["subject"] = subject
message.attach(MIMEText(body, "plain"))
raw_message = base64.urlsafe_b64encode(message.as_bytes()).decode("utf-8")
try:
message = (
service.users()
.messages()
.send(userId="me", body={"raw": raw_message})
.execute()
)
return {"state": True}
except HttpError as error:
print(error)
return {"state": False}
def func(self, mail_dict: dict):
if "action" in mail_dict:
assert mail_dict["action"].lower() in self.__VALID_ACTION__
self.action = mail_dict["action"]
functions = {"read": self._read, "send": self._send}
return functions[self.action](mail_dict)
def convert_action_to(self, action_name: str):
assert (
action_name.lower() in self.__VALID_ACTION__
), f"Action `{action_name}` is not allowed! The valid action is in `{self.__VALID_ACTION__}`"
self.action = action_name.lower()
class WeatherComponet(ToolComponent):
def __init__(self, api_key, name="weather", TIME_FORMAT="%Y-%m-%d"):
super(WeatherComponet, self).__init__(name)
self.name = name
self.TIME_FORMAT = TIME_FORMAT
self.api_key = api_key
def _parse(self, data):
dict_data: dict = {}
for item in data["data"]:
date = item["datetime"]
dict_data[date] = {}
if "weather" in item:
dict_data[date]["description"] = item["weather"]["description"]
mapping = {
"temp": "temperature",
"max_temp": "max_temperature",
"min_temp": "min_temperature",
"precip": "accumulated_precipitation",
}
for key in ["temp", "max_temp", "min_temp", "precip"]:
if key in item:
dict_data[date][mapping[key]] = item[key]
return dict_data
def _query(self, city_name, country_code, start_date, end_date):
"""https://www.weatherbit.io/api/historical-weather-daily"""
# print(datetime.strftime(start_date, self.TIME_FORMAT), datetime.strftime(datetime.now(), self.TIME_FORMAT), end_date, datetime.strftime(datetime.now()+timedelta(days=1), self.TIME_FORMAT))
if start_date == datetime.strftime(
datetime.now(), self.TIME_FORMAT
) and end_date == datetime.strftime(
datetime.now() + timedelta(days=1), self.TIME_FORMAT
):
"""today"""
url = f"https://api.weatherbit.io/v2.0/current?city={city_name}&country={country_code}&key={self.api_key}"
else:
url = f"https://api.weatherbit.io/v2.0/history/daily?&city={city_name}&country={country_code}&start_date={start_date}&end_date={end_date}&key={self.api_key}"
response = requests.get(url)
data = response.json()
return self._parse(data)
def func(self, weather_dict: Dict) -> Dict:
TIME_FORMAT = self.TIME_FORMAT
# Beijing, Shanghai
city_name = weather_dict["city_name"]
# CN, US
country_code = weather_dict["country_code"]
# 2020-02-02
start_date = datetime.strftime(
datetime.strptime(weather_dict["start_date"], self.TIME_FORMAT),
self.TIME_FORMAT,
)
end_date = weather_dict["end_date"] if "end_date" in weather_dict else None
if end_date is None:
end_date = datetime.strftime(
datetime.strptime(start_date, TIME_FORMAT) + timedelta(days=-1),
TIME_FORMAT,
)
else:
end_date = datetime.strftime(
datetime.strptime(weather_dict["end_date"], self.TIME_FORMAT),
self.TIME_FORMAT,
)
if datetime.strptime(start_date, TIME_FORMAT) > datetime.strptime(
end_date, TIME_FORMAT
):
start_date, end_date = end_date, start_date
assert start_date != end_date
return self._query(city_name, country_code, start_date, end_date)
class TranslateComponent(ToolComponent):
__SUPPORT_LANGUAGE__ = [
"af",
"am",
"ar",
"as",
"az",
"ba",
"bg",
"bn",
"bo",
"bs",
"ca",
"cs",
"cy",
"da",
"de",
"dsb",
"dv",
"el",
"en",
"es",
"et",
"eu",
"fa",
"fi",
"fil",
"fj",
"fo",
"fr",
"fr-CA",
"ga",
"gl",
"gom",
"gu",
"ha",
"he",
"hi",
"hr",
"hsb",
"ht",
"hu",
"hy",
"id",
"ig",
"ikt",
"is",
"it",
"iu",
"iu-Latn",
"ja",
"ka",
"kk",
"km",
"kmr",
"kn",
"ko",
"ku",
"ky",
"ln",
"lo",
"lt",
"lug",
"lv",
"lzh",
"mai",
"mg",
"mi",
"mk",
"ml",
"mn-Cyrl",
"mn-Mong",
"mr",
"ms",
"mt",
"mww",
"my",
"nb",
"ne",
"nl",
"nso",
"nya",
"or",
"otq",
"pa",
"pl",
"prs",
"ps",
"pt",
"pt-PT",
"ro",
"ru",
"run",
"rw",
"sd",
"si",
"sk",
"sl",
"sm",
"sn",
"so",
"sq",
"sr-Cyrl",
"sr-Latn",
"st",
"sv",
"sw",
"ta",
"te",
"th",
"ti",
"tk",
"tlh-Latn",
"tlh-Piqd",
"tn",
"to",
"tr",
"tt",
"ty",
"ug",
"uk",
"ur",
"uz",
"vi",
"xh",
"yo",
"yua",
"yue",
"zh-Hans",
"zh-Hant",
"zu",
]
def __init__(
self, api_key, location, default_target_language="zh-cn", name="translate"
):
super(TranslateComponent, self).__init__(name)
self.name = name
self.api_key = api_key
self.location = location
self.default_target_language = default_target_language
def func(self, translate_dict: Dict) -> Dict:
content = translate_dict["content"]
target_language = self.default_target_language
if "target_language" in translate_dict:
target_language = translate_dict["target_language"]
assert (
target_language in self.__SUPPORT_LANGUAGE__
), f"language `{target_language}` is not supported."
endpoint = "https://api.cognitive.microsofttranslator.com"
path = "/translate"
constructed_url = endpoint + path
params = {"api-version": "3.0", "to": target_language}
headers = {
"Ocp-Apim-Subscription-Key": self.api_key,
"Ocp-Apim-Subscription-Region": self.location,
"Content-type": "application/json",
"X-ClientTraceId": str(uuid.uuid4()),
}
body = [{"text": content}]
request = requests.post(
constructed_url, params=params, headers=headers, json=body
)
response = request.json()
response = json.dumps(
response,
sort_keys=True,
ensure_ascii=False,
indent=4,
separators=(",", ": "),
)
response = eval(response)
return {"result": response[0]["translations"][0]["text"]}
class APIComponent(ToolComponent):
def __init__(self):
super(APIComponent, self).__init__()
def func(self, agent) -> Dict:
pass
class FunctionComponent(ToolComponent):
def __init__(
self,
functions,
function_call="auto",
response_type="response",
your_function=None,
):
super().__init__()
self.functions = functions
self.function_call = function_call
self.parameters = {}
self.available_functions = {}
self.response_type = response_type
if your_function:
function_name = your_function["name"]
function_content = your_function["content"]
exec(function_content)
self.available_functions[function_name] = eval(function_name)
for function in self.functions:
self.parameters[function["name"]] = list(
function["parameters"]["properties"].keys()
)
self.available_functions[function["name"]] = eval(function["name"])
def func(self, agent):
messages = agent.long_term_memory
outputdict = {}
query = agent.long_term_memory[-1].content if len(agent.long_term_memory) > 0 else " "
relevant_history = get_relevant_history(
query,
agent.long_term_memory[:-1],
agent.chat_embeddings[:-1],
)
response = agent.LLM.get_response(
messages,
None,
functions=self.functions,
stream=False,
function_call=self.function_call,
relevant_history=relevant_history,
)
response_message = response
if response_message.get("function_call"):
function_name = response_message["function_call"]["name"]
fuction_to_call = self.available_functions[function_name]
function_args = json.loads(response_message["function_call"]["arguments"])
input_args = {}
for args_name in self.parameters[function_name]:
input_args[args_name] = function_args.get(args_name)
function_response = fuction_to_call(**input_args)
if self.response_type == "response":
outputdict["response"] = function_response
elif self.response_type == "prompt":
outputdict["prompt"] = function_response
return outputdict
class CodeComponent(ToolComponent):
def __init__(self, file_name, keyword) -> None:
super().__init__()
self.file_name = file_name
self.keyword = keyword
self.system_prompt = (
"you need to extract the modified code as completely as possible."
)
self.last_prompt = (
f"Please strictly adhere to the following format for outputting: \n"
)
self.last_prompt += (
f"<{self.keyword}> the content you need to extract </{self.keyword}>"
)
def func(self, agent):
response = agent.LLM.get_response(
agent.long_term_memory,
self.system_prompt,
self.last_prompt,
stream=False,
)
code = extract(response, self.keyword)
code = code if code else response
os.makedirs("output_code", exist_ok=True)
file_name = "output_code/" + self.file_name
codes = code.split("\n")
if codes[0] == "```python":
codes.remove(codes[0])
if codes[-1] == "```":
codes.remove(codes[-1])
code = "\n".join(codes)
with open(file_name, "w", encoding="utf-8") as f:
f.write(code)
return {}