init training state

com.unity.ml-agents/Runtime/Policies/TrainingModelRunner.cs

 using Unity.MLAgents.Sensors;
 using UnityEngine;
 using Unity.MLAgents.Inference.Utils;
+using System.Linq;
 
 namespace Unity.MLAgents
 {
         Model m_Model;
         IWorker m_Engine;
         bool m_Verbose = false;
-        string[] m_OutputNames;
         string[] m_TrainingOutputNames;
         IReadOnlyList<TensorProxy> m_TrainingInputs;
         List<TensorProxy> m_TrainingOutputs;
             m_Engine = WorkerFactory.CreateWorker(executionDevice, barracudaModel, m_Verbose);
 
             m_TrainingInputs = barracudaModel.GetTrainingInputTensors();
-            m_OutputNames = barracudaModel.GetOutputNames();
             m_TrainingOutputNames = barracudaModel.GetTrainingOutputNames();
             m_TensorGenerator = new TensorGenerator(
                 seed, m_TensorAllocator, m_Memories, barracudaModel);
 
         void InitializeTrainingState()
         {
-            // TODO: initialize m_TrainingState
+            var initState = m_Model.GetTensorByName(TensorNames.InitialTrainingState);
+            m_TrainingState = new TensorProxy{
+                name = TensorNames.InitialTrainingState,
+                valueType = TensorProxy.TensorType.FloatingPoint,
+                data = initState,
+                shape = initState.shape.ToArray().Select(i => (long)i).ToArray()
+            };
         }
 
         void PrepareBarracudaInputs(IReadOnlyList<TensorProxy> infInputs)
             // Execute the Model
             m_Engine.Execute(m_InputsByName);
 
-            FetchBarracudaOutputs(m_OutputNames);
+            FetchBarracudaOutputs(m_TrainingOutputNames);
 
             // Update the outputs
             m_TensorApplier.ApplyTensors(m_TrainingOutputs, m_OrderedAgentsRequestingDecisions, m_LastActionsReceived);