reduce latest_model prob

config/trainer_config.yaml

     time_horizon: 1000
     self_play:
         window: 10
-        play_against_latest_model_ratio: 0.5
+        play_against_latest_model_ratio: 0.2
         save_steps: 50000
         swap_steps: 50000
         team_change: 100000
     num_layers: 2
     self_play:
         window: 10
-        play_against_latest_model_ratio: 0.5
+        play_against_latest_model_ratio: 0.2
         save_steps: 50000
         swap_steps: 50000
         team_change: 200000