Mapping vs AgentAction

[mroytman83]

I was trying to implement parse function for the handle_parsing_errors parameter in initialize_agent, which is Conversational React, where I would eliminate the json markdown and backticks, and just return the action output key value pair. My return statement for the parse function is {return AgentAction(response["action"], response["action_input"], text)}, and the error that Im receiving is TypeError: langchain.agents.agent.ExceptionTool._run() argument after ** must be a mapping, not AgentAction. Can someone explain what needs to be done? Would be much appreciated!

[dosubot[bot]]

Answer generated by a 🤖

Answer

Hello,

The error you're encountering is due to the fact that the langchain.agents.agent.ExceptionTool._run() method expects a mapping (like a dictionary) as an argument, but you're trying to pass an instance of the AgentAction class.

The AgentAction class in LangChain represents an action that an agent can take. It's typically used in the context of the parse function in the ReActOutputParser class, which parses text into an AgentAction or AgentFinish object. This parse function is designed to handle specific formats of text and convert them into instances of the AgentAction or AgentFinish class, which can then be used in the LangChain framework (source).

However, the langchain.agents.agent.ExceptionTool._run() method expects a dictionary as an argument, not an AgentAction object. This is why you're seeing the TypeError.

To resolve this issue, you should modify your parse function to return a dictionary instead of an AgentAction object. The dictionary should contain the same information as the AgentAction object, but in a format that the langchain.agents.agent.ExceptionTool._run() method can accept.

Here's an example of how you might modify your parse function:

def parse(self, text: str) -> dict:
    action_prefix = "Action: "
    if not text.strip().split("
")[-1].startswith(action_prefix):
        raise OutputParserException(f"Could not parse LLM Output: {text}")
    action_block = text.strip().split("
")[-1]

    action_str = action_block[len(action_prefix) :]
    # Parse out the action and the directive.
    re_matches = re.search(r"(.*?)\[(.*?)\]", action_str)
    if re_matches is None:
        raise OutputParserException(
            f"Could not parse action directive: {action_str}"
        )
    action, action_input = re_matches.group(1), re_matches.group(2)
    if action == "Finish":
        return {"output": action_input}
    else:
        return {"action": action, "action_input": action_input}

This version of the parse function returns a dictionary with the same information as the AgentAction object, but in a format that the langchain.agents.agent.ExceptionTool._run() method can accept.

I hope this helps! Let me know if you have any other questions.

Sources

• libs/langchain/langchain/agents/agent.py
• libs/langchain/langchain/agents/react/output_parser.py

---

This response is meant to be useful, save you time, and share context. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.