How to build LLM-based Classifiers (full tutorial: https://youtu.be/l7NPMiyuh1M?si=bs8V-whzHwdQ63Iv)

evaluate.py

import pandas as pd
from sklearn.metrics import accuracy_score

data = pd.read_csv("data.csv")

print(accuracy_score(data["actual"], data["sentiment"]))

main.py

from openai import OpenAI
import json

client = OpenAI()

product_reviews = [
    "This product is good",
    "The product is not so good",
    "This product is okay",
]

def analyze_sentiment(product_review):
    prompt = f"""
        Analyze the sentiment of the following product review:
        {product_review}
    """

    response = client.chat.completions.create(
        model="gpt-3.5-turbo-1106",
        messages=[{
            "role": "system",
            "content": prompt
        }],
        temperature=0,
        top_p=0.2,
        seed=42,
        tools=[{
            "type": "function",
            "function": {
                "name": "report_sentiment",
                "description": "Report the analyzed sentiment with the explanation for it",
                "parameters": {
                    "type": "object",
                    "properties": {
                        "explanation": {
                            "type": "string",
                            "description": "Explanation of the review sentiment.",
                        },
                        "sentiment": {
                            "type": "string",
                            "enum": ["positive", "negative", "neutral"],
                            "description": "The sentiment of the review..",
                        },
                    },
                    "required": ["explanation", "sentiment"],
                },
            }
        }],
    )

    return json.loads(response.choices[0].message.tool_calls[0].function.arguments)

data = [
    {"review": review, "actual": None, **analyze_sentiment(review)}
    for review in product_reviews
]

import pandas as pd
pd.DataFrame(data).to_csv("data.csv")