Merge commit 'fbcdd83c087135f870e785cc72e5ff9a7e898e3a' into develop-splitpolicyoptimizer

Project/Assets/ML-Agents/Examples/Hallway/Scripts/HallwayAgent.cs

     {
         if (useVectorObs)
         {
-            sensor.AddObservation(GetStepCount() / (float)maxStep);
+            sensor.AddObservation(StepCount / (float)maxStep);
         }
     }
 

com.unity.ml-agents/CHANGELOG.md

  - The stepping logic for the Agent and the Academy has been simplified (#3448)
  - Update Barracuda to 0.6.0-preview
  - The checkpoint file suffix was changed from `.cptk` to `.ckpt` (#3470)
+ - The method `GetStepCount()` on the Agent class has been replaced with the property getter `StepCount`
+- Updated the `gail_config.yaml` to work with per-Agent steps (#3475)
 
 ## [0.14.0-preview] - 2020-02-13
 

com.unity.ml-agents/Runtime/Agent.cs

             m_Brain = m_PolicyFactory.GeneratePolicy(Heuristic);
         }
 
-        /// <summary>
-        /// Current episode number.
+        /// Current step count.
-        public int GetStepCount()
+        public int StepCount
-            return m_StepCount;
+            get { return m_StepCount; }
         }
 
         /// <summary>

com.unity.ml-agents/Tests/Editor/MLAgentsEditModeTest.cs

 
                 Assert.AreEqual(i, aca.TotalStepCount);
 
-                Assert.AreEqual(agent2StepSinceReset, agent2.GetStepCount());
+                Assert.AreEqual(agent2StepSinceReset, agent2.StepCount);
                 Assert.AreEqual(numberAgent1Reset, agent1.agentResetCalls);
                 Assert.AreEqual(numberAgent2Reset, agent2.agentResetCalls);
 
                 expectedAgentStepCount += 1;
 
                 // If the next step will put the agent at maxSteps, we expect it to reset
-                if (agent1.GetStepCount() == maxStep - 1 || (i == 0))
+                if (agent1.StepCount == maxStep - 1 || (i == 0))
-                if (agent1.GetStepCount() == maxStep - 1)
+                if (agent1.StepCount == maxStep - 1)
                 {
                     expectedAgentActionSinceReset = 0;
                     expectedCollectObsCallsSinceReset = 0;
 
-                Assert.AreEqual(expectedAgentStepCount, agent1.GetStepCount());
+                Assert.AreEqual(expectedAgentStepCount, agent1.StepCount);
                 Assert.AreEqual(expectedResets, agent1.agentResetCalls);
                 Assert.AreEqual(expectedAgentAction, agent1.agentActionCalls);
                 Assert.AreEqual(expectedAgentActionSinceReset, agent1.agentActionCallsSinceLastReset);

config/gail_config.yaml

     num_layers: 2
     time_horizon: 64
     sequence_length: 64
-    summary_freq: 1000
+    summary_freq: 10000
     use_recurrent: false
     reward_signals:
         extrinsic:
 Pyramids:
-    summary_freq: 2000
+    summary_freq: 30000
     time_horizon: 128
     batch_size: 128
     buffer_size: 2048
-    max_steps: 5.0e5
+    max_steps: 1.0e7
-        steps: 10000
+        steps: 150000
     reward_signals:
         extrinsic:
             strength: 1.0
     time_horizon: 1000
     batch_size: 2024
     buffer_size: 20240
-    max_steps: 1e6
-    summary_freq: 3000
+    max_steps: 1e7
+    summary_freq: 30000
-        steps: 5000
+        steps: 50000
     reward_signals:
         gail:
             strength: 1.0
 
 PushBlock:
-    max_steps: 5.0e4
+    max_steps: 1.5e7
-    summary_freq: 2000
+    summary_freq: 60000
     time_horizon: 64
     num_layers: 2
     reward_signals:
             encoding_size: 128
-            demo_path: Project/Assets/ML-Agents/Examples/PushBlock/Demos/ExpertPush.demo
+            demo_path: Project/Assets/Demonstrations/PushblockDemo.demo
 
 Hallway:
     use_recurrent: true
     num_epoch: 3
     buffer_size: 1024
     batch_size: 128
-    max_steps: 5.0e5
-    summary_freq: 1000
+    max_steps: 1.0e7
+    summary_freq: 10000
     time_horizon: 64
     reward_signals:
         extrinsic:
 
 FoodCollector:
     batch_size: 64
-    summary_freq: 1000
-    max_steps: 5.0e4
+    max_steps: 2.0e6
     use_recurrent: false
     hidden_units: 128
     learning_rate: 3.0e-4

docs/Migrating.md

 * The `Monitor` class has been moved to the Examples Project. (It was prone to errors during testing)
 * The `MLAgents.Sensor` namespace has been removed. All sensors now belong to the `MLAgents` namespace.
 * The `SetActionMask` method must now be called on the optional `ActionMasker` argument of the `CollectObservations` method. (We now consider an action mask as a type of observation)
+* The method `GetStepCount()` on the Agent class has been replaced with the property getter `StepCount`
+* Replace all calls to `Agent.GetStepCount()` with `Agent.StepCount`
 
 ## Migrating from 0.13 to 0.14