PMLDL Assignment 1 – MNIST Deployment

This repository contains a minimal solution for Assignment 1 in the PMLDL course.
The goal is to prepare data, train a simple CNN on MNIST, expose it via a FastAPI backend, and build a small Streamlit frontend to draw digits and get predictions.

---

📂 Project structure

.
├── docker-compose.yml
├── README.md
└── src
    ├── api
    │   ├── api.py
    │   ├── Dockerfile
    │   └── requirements.txt
    ├── app
    │   ├── app.py
    │   ├── Dockerfile
    │   └── requirements.txt
    └── prepare
        ├── dataset
        │   └── data_processing.py
        ├── Dockerfile
        ├── model
        │   └── model.py
        └── requirements.txt

---

🚀 How to run (with Docker)

1. Build and start all services:

   docker compose up --build

2. Workflow:

   • prepare: downloads MNIST, saves train.pt / test.pt, trains CNN, and saves weights to simple_nn.pth.

   • api: loads the trained model and serves endpoints:

     • GET /health → returns {"status": "ok"}
     • POST /predict → takes a 28×28 digit array and returns prediction + probabilities.

   • app: launches Streamlit UI to draw digits and call the API.

3. Open the frontend in your browser: http://localhost:8501