Stock Price Prediction API Server (Jupyter Notebook)

This project provides a simple FastAPI server that performs stock price prediction using a Transformer-based model. It is designed to run within a Jupyter Notebook environment and allows external API access through ngrok.

Key Features

• Loads a 4-bit quantized stock prediction model (finma-7b)
• Runs a REST API server using FastAPI
• Generates an external access URL via ngrok

Required Libraries

The following libraries should be installed within the notebook:

pip install fastapi uvicorn nest-asyncio pyngrok
pip install transformers accelerate torch pandas yfinance finance-datareader
pip install bitsandbytes peft shap matplotlib

How to Run

1. Open the API_server.ipynb file and run each cell sequentially.
2. Load the model → Start the server → Display the ngrok URL
3. Use the displayed ngrok URL to make API calls from outside the notebook.

Example Request

GET /predict?ticker=005930.KS&horizon_days=7

• ticker: Stock symbol (e.g., AAPL, 005930.KS)
• horizon_days: Prediction period (choose from 1, 7, or 30 days)

Notes

• Requires at least 6GB of GPU RAM. Not compatible with EC2 free tier; assumed to run on Google Colab.
• For production deployment, consider converting to a standalone Python script or Docker container.
• Registration of Hugging Face and ngrok tokens is necessary.