langchain: Toolkits - Pandas Dataframe Agent failed to call "python_repl_ast" consistently

Please ask: how many rows

Entering new AgentExecutor chain… Thought: I need to find the number of rows in the dataframe Action: python_repl_ast Action Input: len(df) Observation: 45 Thought:I have the number of rows Final Answer: 45

Finished chain. 45

Please ask: does fg-40f support local report

Entering new AgentExecutor chain… Thought: We need to check if the “FG-40F Series” row has “Yes” in the “Local Reporting” column. Action: Use pandas boolean indexing to filter the dataframe to only the “FG-40F Series” row and check the value in the “Local Reporting” column. Action Input: df[df['Product'] == 'FG-40F Series']['Local Reporting'] Observation: Use pandas boolean indexing to filter the dataframe to only the “FG-40F Series” row and check the value in the “Local Reporting” column. is not a valid tool, try another one. Thought:I made a mistake, I should use python_repl_ast instead of Observation. Action: Use python_repl_ast to execute the previous input. Action Input: python_repl_ast(df[df['Product'] == 'FG-40F Series']['Local Reporting']) Observation: Use python_repl_ast to execute the previous input. is not a valid tool, try another one. Thought:I made another mistake, I should use print() to display the result. Action: Use python_repl_ast to execute the previous input with print(). Action Input: python_repl_ast(print(df[df['Product'] == 'FG-40F Series']['Local Reporting']))

About this issue

Original URL
State: open
Created a year ago
Comments: 15 (3 by maintainers)

Commits related to this issue

Fixed Ast Python Repl for Chatgpt multiline commands (#2406) Resolves issue https://github.com/hwchase17/langchain/issues/2252 --------- Co-authored-by: Abhik Singla <abhiksingla@microsoft.com> — committed to langchain-ai/langchain by abhiksingla a year ago

Most upvoted comments

@MohammedAlhajji I am kind of making a hardcode solution working with ChatGPT-3.5-turbo. It is not good for general purposes but just want to share some ideas with you.

I had to make a clone of ‘create_pandas_dataframe_agent’ function in my own code so I can pass my FORMAT_INSTRUCTIONS to ZeroShotAgent.create_prompt
My FORMAT_INSTRUCTIONS is set as

FORMAT_INSTRUCTIONS = """Use the following format:

Question: the input question you must answer
Thought: you should always think about what to do, what action to take
Action: python_repl_ast
Action Input: the input to the action, never add backticks "`" around the action input
Observation: the result of the action
... (this Thought/Action/Action Input/Observation can repeat N times)
Thought: I now know the final answer
Final Answer: the final answer to the original input question"""

Action: python_repl_ast to force LLM to pick python_repl_ast as the tool Action Input: the input to the action, never add backticks “`” around the action input to tell LLM NOT to add backticks

zhangjw71 on Apr 2, 2023

Using agent_type=AgentType.OPENAI_FUNCTIONS with gpt-3.5-turbo might fix the issue

the13thdev on Sep 27, 2023

Hi! Any update on the fix? The agent (using llm gpt-35-turbo) goes into a loop until max iterations is reached and throws error “Agent stopped due to iteration limit or time limit”.

satwikide on Sep 6, 2023

If anyone visits here because of the python is not a valid tool, try one of [python_repl_ast].

Try to add if function_name == "python": function_name = "python_repl_ast" to lib/site-packages/langchain/agents/ouput_parsers/openai_functions.py before the content_msg where its at approximately 70th line.

yusufdemyr on Nov 3, 2023

return_direct = True，response就可以获取大模型回答的结果了

xxm1668 on Oct 12, 2023

By changing Action: python_repl_ast to Action: the tool name which should be “python_repl_ast”

And clone my own PythonAstREPLTool to ignore surrounded backticks if any.

I can get the Python tool running quite stably. However, ChatGPT3.5-turbo failed to generate good python code consistently.

zhangjw71 on Apr 2, 2023