NeuroDraughts

Red Black

Press Start training, and a game from each batch replays here.

Replay speed

Look

Training

Network

Settings

Learning rate Halve it every (games, 0 for never) Random moves (exploration) Trace decay (lambda) Training games per batch Test games per batch Score needed to clone Train against the last (champions) Look-ahead (plies) Seed

Predict win, draw and loss chances (not one number) Learn from both players' positions Learn only from quiet positions (no capture pending)

Changes to these apply from the next batch. The network and seed are fixed once a run starts.

Generation0

Training games0

Games per second-

Batch0

Evolution

Each batch the learner plays its champion, the last clone of itself. When it scores enough in the test it is cloned, and its reward is to have to beat itself again. Numbered lines mark each clone.

Generations

Every champion is kept. Each one plays 40 games against the network I shipped in June 1997, which has no idea what is coming.

Gen	After	Against the 1997 champion

No generations yet.

You against the networks

Everything you have played

Against	It looked ahead	Won	Drew	Lost

Nothing yet. Games count once they are played to the end.

Everyone against the networks

Leaderboard

Arena

Pit any two players against each other, or bring in a network file someone sent you.

against

Import a network file