Langchain and Chroma Parse CSV and embed into ChatGPT not returning proper responses

Question

I am trying to parse a Stardew Valley CSV, embed that into ChaptGPT, and have ChatGPT answer questions about the data. The problem is that my responses I get from ChatGPT are not accurate. I am still learning to use langchain and Chroma so I am unsure if there is something wrong with my code or its on my end.

Question	Response	Correct
What sells for the most between a pumpkin or rare seed?	The rare seed sells for more than the pumpkin	Yes
What crop sell for them most in the fall?	The crop that sells for the most in Fall is the pumpkin, with a selling price of 320	No
What are all the crops in the fall?	The crops in fall are Corn, Pumpkin, and Cranberry.	No (there are more)

import os
import sys
import constants

# import langchain loaders
from langchain_openai import ChatOpenAI
from langchain_community.document_loaders.csv_loader import CSVLoader
from langchain_openai import OpenAIEmbeddings
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnableLambda, RunnablePassthrough
from langchain_community.vectorstores import Chroma

os.environ["OPENAI_API_KEY"] = constants.OPENAI_API_KEY

embedding_function = OpenAIEmbeddings()

loader = CSVLoader(file_path="./stardew-crops.csv", csv_args={
    'delimiter': ',',
    'fieldnames': ['Season', 'Crop', 'Buy', 'Sell', 'Grow', 'Regr', 'Yield', 'Profit/D', 'Profit/M', 'Vendor', 'Notes']
})
documents = loader.load()

db = Chroma.from_documents(documents, embedding_function)
retriever = db.as_retriever()

template = """You are a Stardew Valley crop specialist. 
{context}

Question: {question}
"""
prompt = ChatPromptTemplate.from_template(template)
model = ChatOpenAI()

chain = (
    {"context": retriever, "question": RunnablePassthrough()}
    | prompt
    | model
    | StrOutputParser()
)

# Continuous querying loop
while True:
    query = input("Please enter your query (or type 'exit' to quit): ")
    if query.lower() == 'exit':
        print("Exiting...")
        break

    # Invoke the chain with the current query and print the response
    response = chain.invoke(query)
    print(response)

Here are a few rows from the CSV:

Fall,Fairy Rose,200,290,12,0,1.0,7.50,180,Pierre,
Fall,Grape,60,80,10,3,1.0,2.00,420,Pierre,
Fall,Pumpkin,100,320,13,0,1.0,16.92,440,Pierre,
Fall,Rare Seed,1000,3000,24,0,1.0,83.33,2000,Gypsy,Receive Sweet Gem Berry

Leave a Comment Cancel reply