Langchain and prompting in dataframes (not with create_pandas_datafame)

[tlallema]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

from langchain.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_core.messages import HumanMessage

query = f'''SELECT * FROM table_name'''
df = connectToDB().query_simple(query)
df_5_rows=df.head()
csv_string=df_5_rows.to_csv(index=False)
prompt = ChatPromptTemplate.from_messages(
        [
            (
                "system","""
                you are a data vizualisation expert and use your favourite graphing library Plotly only. 
                You are working with a pandas dataframe in Python.
                Here are the first 5 rows of the data set : {data}
                Follow the user's indication when creating the chart"""
            ),
            MessagesPlaceholder(variable_name="messages")
        ]
    )
chain = prompt | get_openai_client()
response=chain.invoke({
        "messages":[HumanMessage(content=user_input)],
        "data": csv_string},
    )
result_output = response.content

Description

I'm trying to build a prompt with langchain which could use dataframes (here dataframe from sql tables). This prompt works fine when giving name of files (csv or xls). It returns chart based on the underlying data (using Plotly library). However I do no succeed to do it with pandas dataframes. any idea how I should modify the context/system in the prompt (not going through the Create_pandas_dataframe) ?

System Info

last langchain version, python 3.9

[dosubot[bot]]

Hey @tlallema! I'm here to help you with any issues you're facing. Let's figure this out together!

To modify the context/system in the LangChain prompt to work directly with pandas dataframes without converting them to CSV strings, you can use the PandasDataFrameOutputParser to handle the dataframe directly. Here is an example of how to do this:

import pandas as pd
from langchain_core.prompts import ChatPromptTemplate
from langchain_experimental.tools import PythonAstREPLTool
from langchain.output_parsers.pandas_dataframe import PandasDataFrameOutputParser
from langchain_openai import ChatOpenAI

# Load your dataframe
df = pd.read_csv("titanic.csv")

# Create the tool with the dataframe in its local scope
tool = PythonAstREPLTool(locals={"df": df})

# Create the system prompt
system = f"""You have access to a pandas dataframe `df`. \
Here is the output of `df.head().to_markdown()`:

{df.head().to_markdown()}

Given a user question, write the Python code to answer it. \
Return ONLY the valid Python code and nothing else. \
Don't assume you have access to any libraries other than built-in Python ones and pandas."""

# Create the prompt template
prompt = ChatPromptTemplate.from_messages([("system", system), ("human", "{question}")])

# Create the LLM with tools and parser
llm = ChatOpenAI(model="gpt-3.5-turbo", temperature=0)
llm_with_tools = llm.bind_tools([tool], tool_choice=tool.name)
parser = PandasDataFrameOutputParser(dataframe=df)

# Combine the prompt, LLM, parser, and tool into a chain
chain = prompt | llm_with_tools | parser | tool

# Invoke the chain with a question
result = chain.invoke({"question": "What's the correlation between age and fare"})
print(result)

This setup allows the LangChain prompt to work directly with pandas dataframes by including the dataframe's head in the system prompt and using the PandasDataFrameOutputParser to handle the dataframe operations. Note that using a library like Pandas requires letting the model execute Python code, which carries significant security risks. For production use cases, it is recommended to load the CSV(s) into a SQL database and use SQL queries instead ^[1][2][3].

To continue talking to Dosu, mention @dosu.

[tlallema]

@dosu
copy/pasting exactly your code and still getting the same error.
can someone else than the bot help ?

[dosubot[bot]]

@eyurtsev, this one's a bit out of my wheelhouse, could you step in and provide your guidance?

To continue talking to Dosu, mention @dosu.

[tlallema]

@eyurtsev
any feedback would be very appreciated.
Thanks a lot

[tlallema]

hello @dosu, @eyurtsev ,
could you please come back to me with a solution to this issue ?

[dosubot[bot]]

A maintainer has already been looped in.

To continue talking to Dosu, mention @dosu.