# EECE571F project: Reinforcement Learning for Variable Selection in Branch and Bound

## About
This repo is largely un-organized and I wouldn't suggest running any of the code unless you read through this document.

## Acknowledgement
A portion of our code to implement Exact Combinatorial Optimization with Graph Convolutional Neural Networks, Gasse et. all, is derived from [ml4co-competition](https://github.com/ds4dm/ml4co-competition). 

## Description 
Our final code to produce results is stored in `ml4co-competition/common` and `ml4co-competition/submissions`. Earlier versions of the project code is stored in the following folders:
- `B&B` --> Version 1
- `learn2branch_gasse` --> Version 2; uses Ecole.ai
- `learn2branch_self` --> Version 3; first attempt to solve variable selection with Ecole.ai and RL using an on-policy method
- `ml4co-competition` --> Version 4; final code to reproduce results, uses Ecole.ai and an off-policy RL agent.

An excel file is in the root of the repo, which was used to calculate performance metrics outlined in our report.