new soccertwos brain

config/trainer_config.yaml

 
 SoccerTwos:
     normalize: false
-    max_steps: 1.0e8
+    max_steps: 5.0e7
     learning_rate_schedule: constant
     batch_size: 2048
     buffer_size: 20480
     self_play:
-        window: 50
+        window: 10
         play_against_latest_model_ratio: 0.5
         save_steps: 50000
         swap_steps: 50000