Can we create agents that practice self-optimizing for Balatro?

[TomLucidor]

There are multiple repos out there which claims to evolve code based on standard metrics or tasks, and I wonder if Balatro can be made to fit the bill https://github.com/codelion/openevolve https://github.com/jennyzzt/dgm

The only problem is that:

• this game is much harder to map than classical card games (RL-Card is multi-agent so we also need to consider single-agent or "solitare" games as well) https://github.com/datamllab/rlcard
• There are parts worthy of deep learning, and other parts worthy of using hard-coded rules, and then some that are neuro-symbolic
• What is the compute budget again?