src/E_openAI_model.py

import os
import pandas as pd
from openai import OpenAI
import streamlit as st
from dotenv import load_dotenv
import re

load_dotenv()  # take environment variables from .env.
api_key = os.getenv('OPENAI_API_KEY')
#print(api_key)
# Create a filtered dataframe with the most 20 ballanced intents
def get_ballanced_intents(data):
    intents = data['intent'].value_counts()
    intents = intents[intents <= 40].index.tolist()
    filtered_data = data[data['intent'].isin(intents)]
    print(f"Filtered data shape: {filtered_data.shape}")
    return filtered_data
  



# Create a prompt for the OpenAI Chat model
def create_prompt(user_text, utterances):
    prompt = f"User text: {user_text}\n\nUtterance examples:\n"
    for i, utterance in enumerate(utterances):
        prompt += f"{i + 1}. {utterance}\n"
    prompt += "\nPlease rate the similarity of the user text to each of the utterance examples on a scale from 0 to 1."
    return prompt

# Get the similarity scores for the user text and the utterance examples
def get_similarity_scores(prompt, temperature):
    load_dotenv()  # take environment variables from .env.
    api_key = os.getenv('OPENAI_API_KEY')
    #print(api_key)
    if not api_key:
        st.write("Please set the OPENAI_API_KEY environment variable.")
        return

    #client = OpenAI(api_key=api_key)
    client = OpenAI(api_key='sk-proj-SNnm3Z9t6BvJgQ3ztjWLT3BlbkFJ1037ZT68ltN47zcir44l')
    response = client.chat.completions.create(
        model="gpt-4o",
        messages=[
            {"role": "system", "content": "You are a helpful assistant. Respond without details in ()"},
            {"role": "user", "content": prompt}
        ],
        max_tokens=150,
        n=1,
        stop='8.',
        temperature=temperature,
    )

    similarities = response.choices[0].message.content.strip().split('\n')
    #print(similarities)
    for sim in similarities:
        #print(sim)
        print(sim.split('- ')[-1])
    similarity_scores = [sim.split('- ')[-1] for sim in similarities]
    similarity_scores = [sim if re.match(r'^0\.[1-9]$|^1\.0$', sim) else '0' for sim in similarity_scores] # Filter out the similarity scores

    return similarity_scores

# Get the most similar intent and its confidence score
def get_most_similar_intent(user_text, utterances, intents, temperature=0.5):
    prompt = create_prompt(user_text, utterances)
    similarity_scores = get_similarity_scores(prompt, temperature)
    print(similarity_scores)
    max_index = similarity_scores.index(max(similarity_scores))
    print(f'max_index: {max_index}')
    most_similar_intent = intents[max_index]
    
    return most_similar_intent, similarity_scores[max_index]