RL: self-play PPO trening siete

BridzikNet (trup + guess/play/value hlavy s maskovanim), self-play generator so zdielanou sietou na 4 sedadlach a opponent mixingom (random/heuristicke sedadla pre robustnost), vlastny clipped-PPO so skalovanim odmien a lr/entropy annealom. Torch je len trenovacia zavislost na hoste (requirements-rl.txt); checkpointy a logy su gitignorovane. Spustenie: py -m rl.train. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-07 18:49:55 +02:00
parent e1733f4943
commit 8f2449a408
7 changed files with 659 additions and 0 deletions
@@ -0,0 +1,5 @@
+# RL trening (rl/model.py, rl/selfplay.py, rl/train.py) -- zamerne oddelene
+# od requirements.txt: server ani Docker image torch nepotrebuju, boti v hre
+# pouzivaju len cisto-Python rl/players.py (a neskor natrenovane vahy cez
+# torch az ked sa neuralny bot nasadi).
+torch>=2.4