In LangChain, how do I use AgentExecutor to replace the disabled initialize_agent function?

Alternative to initialize_agent function in LangChain: Use AgentExecutor

LangChain's initialize_agent function has been deprecated, and it is recommended to use a more flexible AgentExecutor class to initialize and run the agent. AgentExecutor provides finer control and customization options. Here are the steps to use AgentExecutor :

1. Import the necessary modules:

 from langchain.agents import AgentExecutor
from langchain.agents.tools import Tool

2. Definition Tools:

Agents need a range of tools to complete tasks. These tools can be functions, API calls, or other executable units. For example:

 def search_tool(query):
    # Implement your search logic here, for example using SerpAPI or other search engines#...
    return "Search results"

tools = [
    Tool(
        name="Search",
        func=search_tool,
        description="Tool for searching information, enter query keywords."
    )
]

3. Create an agent:

Choose the appropriate proxy type, such as ZeroShotAgent or others. The following example uses ZeroShotAgent :

 from langchain.agents import ZeroShotAgent
from langchain import LLMChain, SerpAPIWrapper # Replace with your LLM and search tools # Replace with your LLM
llm = ... 

search = SerpAPIWrapper() # or other search tools = [
    Tool(
        name="Calculator",
        func=lambda x: str(eval(x)), #Simple calculator example, more complete processing description="Tool for simple mathematical calculations, enter mathematical expressions."
    ),
    Tool(
        name="Search",
        func=search.run,
        description="Tool for searching information, enter query keywords."
    )
]

prompt = ZeroShotAgent.create_prompt(
    tools,
    prefix="Please answer the question according to the following tools:",
    suffix="start!"
)

llm_chain = LLMChain(llm=llm, prompt=prompt)
agent = ZeroShotAgent(llm_chain=llm_chain, tools=tools)

4. Initialize and use AgentExecutor :

 agent_executor = AgentExecutor.from_agent_and_tools(agent=agent, tools=tools, verbose=True)

response = agent_executor.run("What is the population of Tokyo?")
print(response)

Through the above steps, you can use AgentExecutor to replace the disabled initialize_agent function and manage and run your LangChain agent more efficiently. Remember to replace the placeholder ( ... ) in the sample code for the LLM and tool you are actually using. verbose=True will print the reasoning process of the proxy, which is convenient for debugging.

The above is the detailed content of In LangChain, how do I use AgentExecutor to replace the disabled initialize_agent function?. For more information, please follow other related articles on the PHP Chinese website!