🎒 Knapsack Autoencoder API - FastAPI + Keras 🚀

Encode and decode variable-sized 0/1 Knapsack Problem instances (from 50 to 1000 items) using a state-of-the-art Autoencoder served via FastAPI.

This project combines the flexibility of a custom-trained Keras Autoencoder, built with exhaustive keras_tuner optimization and learning rate scheduling, with the speed and simplicity of a FastAPI backend. It even supports reading problem instances directly from the file system.

---

✨ Features

• Autoencoder:

  • 🧠 Variable Size Autoencoder trained for knapsack instances with 50–1000 items.
  • 🔍 Exhaustive hyperparameter tuning via keras_tuner for optimal architecture.
  • 📉 Dynamic learning rate scheduling for better convergence.

• ⚡ FastAPI-powered API for quick encode/decode operations.

• 📁 Filesystem support to read and process instance files on the fly.

• 🧪 Ready for integration into optimization pipelines or benchmarking tools.

• Docker container available

---

🔧 Endpoints

POST /encode

Encode a collection of knapsack instances to its latent representation.

{
  "collection": {
    "instances": []
  },
  "encodings": [
    [
      9.996188163757324,
      14.908147811889648
    ]
  ]
}

POST /decode

Decode from a latent vector back to an approximate knapsack instance.

POST /encode/filename

Upload a file (e.g., .kp) containing the knapsack instance. The expected structure of the file is:

N Q

w0 p0
w1 p1
...
wN-1 pN-1

---

🚀 Getting Started

Clone the repo

git clone https://github.com/yourusername/knapsack-autoencoder-api.git
cd knapsack-ae-api

Run the API

uv run fastapi run app/app.py

Run using Docker

docker run -d -p 8000:80 amarrerd/knapsack_ae_api:0.0.1

---

🧠 Model Info

• Encoder/decoder trained using Keras Functional API.
• Latent vector size 2d.
• Trained dataset includes 100,000+ synthetic knapsack instances.

---

📁 Example Files

Use the /test folder for sample knapsack problem files in supported formats.

---

📬 Contact

Feel free to open an issue or PR! Contributions and suggestions are welcome.

---