Reinforcement Learning Solver

[knagrecha]

Summary

Currently, the only supported method of solving the SPASE problem is to use an MILP solver like Gurobi. Gurobi licenses are free for academic users but can be pretty expensive for everyone else. Open-source MILP solvers do not perform as well. I propose that we make a reinforcement-learning-based solver as an open-source alternative for users without access to Gurobi.

Key Proposal
A new RL solver for Saturn's SPASE problem.

What it would take
This would be a significant overhaul. RL is not as reliable as an MILP solver. We'd need a lot of testing to be sure this works.

@knagrecha @knagrecha-nflx