Supervised Training

Train a model from recorded gameplay and activate it for live play.

Decision Tree: quick baseline for tabular patterns.

MLP-specific settings are shown only when MLP is selected.

How are "samples from rounds" calculated?

Each game session with n rounds contributes max(0, n - lookback) samples.

Checking training readiness...

Job Status

No active training job.


  

Benchmark Against Canonical Bots

Target: non-tie win rate of at least 0.60 across quincy, abbey, kris, and mrugesh.

API benchmarks use a 20-second runtime budget to avoid request timeouts. Use CLI for longer runs.

No benchmark run yet.

Model Registry

ID Name Type Test Accuracy Active Action