|
|
|
|
|
|
by eclark
2 days ago
|
|
|
No this doesn't currently use a model for valuation. It uses recursive exploration(with a stockfish like early terminition) and keeps track of the reward and regret at each node. With brute force on number of cores and compact memory layout I was able to get reasonable convergence with 6 way rings. I do have some plans to release a follow on with some ml components ala Pluribus. |
|
|