app.py

import os
import shutil
import gradio as gr
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
from transformers import pipeline

# Initialize Hugging Face Chat Model (Open-source LLM)
chatbot_pipeline = pipeline("text-generation", model="mistralai/Mistral-7B-Instruct-v0.1")

base_prompt = """You are an expert data analyst.
Analyze the dataset structure and determine the best target variable.
List 3 interesting questions about correlations in the data.
Answer these questions with relevant numbers and real-world insights.
Generate relevant plots using Matplotlib/Seaborn and save them to './figures/'.
Ensure each figure is cleared before creating another.
Structure of the dataset:
{structure_notes}
The data is already loaded as a pandas dataframe named `data_file`.
"""

def get_images_in_directory(directory):
    image_extensions = {'.png', '.jpg', '.jpeg', '.gif', '.bmp', '.tiff'}
    return [os.path.join(directory, file) for file in os.listdir(directory) if os.path.splitext(file)[1].lower() in image_extensions]

def interact_with_agent(file_input, additional_notes):
    shutil.rmtree("./figures", ignore_errors=True)
    os.makedirs("./figures", exist_ok=True)

    data_file = pd.read_csv(file_input)
    data_structure_notes = f"""- Description:
    {data_file.describe()}
    - Columns and types:
    {data_file.dtypes}"""

    prompt = base_prompt.format(structure_notes=data_structure_notes)
    if additional_notes:
        prompt += "\nAdditional Notes:\n" + additional_notes

    yield [gr.ChatMessage(role="assistant", content="⏳ _Analyzing dataset..._")]

    # Generate response using Hugging Face LLM
    response = chatbot_pipeline(prompt, max_length=1024, do_sample=True)[0]['generated_text']
    
    messages = [{"role": "user", "content": prompt}, {"role": "assistant", "content": response}]
    
    # Placeholder for visualization (if required)
    for image_path in get_images_in_directory("./figures"):
        messages.append(gr.ChatMessage(role="assistant", content=gr.FileData(path=image_path, mime_type="image/png")))

    yield messages

# Gradio UI for Hugging Face Spaces
demo = gr.Blocks()

with demo:
    gr.Markdown("# GPT Data Analyst (Hugging Face) 📊🤖")
    file_input = gr.File(label="Upload CSV file")
    text_input = gr.Textbox(label="Additional notes")
    submit = gr.Button("Run Analysis!", variant="primary")
    chatbot = gr.Chatbot(label="Data Analyst Assistant", type="messages")
    submit.click(interact_with_agent, [file_input, text_input], [chatbot])

if __name__ == "__main__":
    demo.launch(share=True)  # Enable public sharing on HF Spaces