Update selfchess.py

selfchess.py

@@ -16,7 +16,7 @@ CONFIG = {
     "num_games": 3000,
     "stockfish_time_limit": 1.0,
     "search_depth": 1,
-    "epsilon": 0.2
+    "epsilon": 4
 }
 
 device = CONFIG["device"]
@@ -157,8 +157,11 @@ def game_gen(engine_side):
             keys = list(evaling.keys())
             logits = torch.tensor(list(evaling.values())).to(device)
             probs = torch.softmax(logits,dim=0)
-            idx = torch.multinomial(probs,num_samples=1)
-            move = keys[idx]
+            bests = torch.multinomial(probs,num_samples=CONFIG["epsilon"])
+            bestl = []
+            for idx in bests:
+                bestl += [logits[idx].item()]
+            move = keys[torch.argmax(bestl).item()]
         else:
             result = engine.play(board, lim)
             move = result.move