tennis reward fix

Project/Assets/ML-Agents/Examples/Tennis/Prefabs/TennisArea.prefab

   myArea: {fileID: 1541947554534326}
   angle: 0
   scale: 0
+  timePenalty: 0
 --- !u!114 &2449890524009497851
 MonoBehaviour:
   m_ObjectHideFlags: 0
   myArea: {fileID: 1541947554534326}
   angle: 0
   scale: 0
+  timePenalty: 0
 --- !u!114 &6598495797138489682
 MonoBehaviour:
   m_ObjectHideFlags: 0

Project/Assets/ML-Agents/Examples/Tennis/Scripts/HitWall.cs

     
     void AgentAWins()
     {
-        m_AgentA.SetReward(1);
+        m_AgentA.SetReward(1 + m_AgentA.timePenalty);
         m_AgentB.SetReward(-1);
         m_AgentA.score += 1;
         Reset();
     void AgentBWins()
     {
         m_AgentA.SetReward(-1);
-        m_AgentB.SetReward(1);
+        m_AgentB.SetReward(1 + m_AgentB.timePenalty);
         m_AgentB.score += 1;
         Reset();
 

Project/Assets/ML-Agents/Examples/Tennis/Scripts/TennisAgent.cs

     public float angle;
     public float scale;
 
+    [HideInInspector]
+    public float timePenalty = 0;
+
     Text m_TextComponent;
     Rigidbody m_AgentRb;
     Rigidbody m_BallRb;
                 transform.position.y,
                 transform.position.z);
         }
-        AddReward(-1f / 3000f);
+       // AddReward(-1f / 3000f);
+       timePenalty += -1f / 3000f;
 
         m_TextComponent.text = score.ToString();
     }
     public override void OnEpisodeBegin()
     {
 
+        timePenalty = 0;
         m_InvertMult = invertX ? -1f : 1f;
 
         transform.position = new Vector3(-m_InvertMult * Random.Range(6f, 8f), -1.5f, -1.8f) + transform.parent.transform.position;

config/trainer_config.yaml

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