Cochon123/clash-royale-policy-bc-v4-1
0
Clash Royale policy BC v4.1
Same architecture as v4.0 (card-conditioned placement), retrained on a larger replay cut.
Summary
- Trained on: 28,102 usable battles (598,192 train samples), best checkpoint at epoch 17/25
- Created: 2026-08-08
- Compare report:
reports/policy_bc_v4_1_compare.html
Note
The training process was killed by memory pressure before finishing all 25 epochs. Artifacts are from the best validation checkpoint (epoch 17).
