public void InitializeCoreBrain(MLAgents.Batcher brainBatcher)
{
Debug.Assert(trainer != null && trainerInterface != null, "Please specify a trainer in the Trainer field of your Brain!");
trainerInterface.Initialize();
}
/// <summary>
/// Initializes the environment, configures it and initialized the Academy.
/// </summary>
private void InitializeEnvironment()
{
m_OriginalGravity = Physics.gravity;
m_OriginalFixedDeltaTime = Time.fixedDeltaTime;
m_OriginalMaximumDeltaTime = Time.maximumDeltaTime;
InitializeAcademy();
ICommunicator communicator;
var exposedBrains = broadcastHub.broadcastingBrains.Where(x => x != null).ToList();
var controlledBrains = broadcastHub.broadcastingBrains.Where(
x => x != null && x is LearningBrain && broadcastHub.IsControlled(x));
foreach (var brain1 in controlledBrains)
{
var brain = (LearningBrain)brain1;
brain.SetToControlledExternally();
}
// Try to launch the communicator by usig the arguments passed at launch
try
{
communicator = new RpcCommunicator(
new CommunicatorParameters
{
port = ReadArgs()
});
}
// If it fails, we check if there are any external brains in the scene
// If there are : Launch the communicator on the default port
// If there arn't, there is no need for a communicator and it is set
// to null
catch
{
communicator = null;
if (controlledBrains.ToList().Count > 0)
{
communicator = new RpcCommunicator(
new CommunicatorParameters
{
port = 5005
});
}
}
m_BrainBatcher = new Batcher(communicator);
foreach (var trainingBrain in exposedBrains)
{
trainingBrain.SetBatcher(m_BrainBatcher);
}
if (communicator != null)
{
m_IsCommunicatorOn = true;
var academyParameters =
new CommunicatorObjects.UnityRLInitializationOutput();
academyParameters.Name = gameObject.name;
academyParameters.Version = k_KApiVersion;
foreach (var brain in exposedBrains)
{
var bp = brain.brainParameters;
academyParameters.BrainParameters.Add(
bp.ToProto(brain.name, broadcastHub.IsControlled(brain)));
}
academyParameters.EnvironmentParameters =
new CommunicatorObjects.EnvironmentParametersProto();
foreach (var key in resetParameters.Keys)
{
academyParameters.EnvironmentParameters.FloatParameters.Add(
key, resetParameters[key]
);
}
var pythonParameters = m_BrainBatcher.SendAcademyParameters(academyParameters);
Random.InitState(pythonParameters.Seed);
}
// If a communicator is enabled/provided, then we assume we are in
// training mode. In the absence of a communicator, we assume we are
// in inference mode.
m_IsInference = !m_IsCommunicatorOn;
BrainDecideAction += () => { };
DestroyAction += () => { };
AgentSetStatus += (i) => { };
AgentResetIfDone += () => { };
AgentSendState += () => { };
AgentAct += () => { };
AgentForceReset += () => { };
// Configure the environment using the configurations provided by
// the developer in the Editor.
SetIsInference(!m_BrainBatcher.GetIsTraining());
ConfigureEnvironment();
}
/// Loads the tensorflow graph model to generate a TFGraph object
public void InitializeCoreBrain(MLAgents.Batcher brainBatcher)
{
#if ENABLE_TENSORFLOW
#if UNITY_ANDROID && !UNITY_EDITOR
// This needs to ba called only once and will raise an exception if
// there are multiple internal brains
try{
TensorFlowSharp.Android.NativeBinding.Init();
}
catch {
}
#endif
if ((brainBatcher == null) ||
(!broadcast))
{
this.brainBatcher = null;
}
else
{
this.brainBatcher = brainBatcher;
this.brainBatcher.SubscribeBrain(brain.gameObject.name);
}
if (graphModel != null)
{
graph = new TFGraph();
graph.Import(graphModel.bytes);
session = new TFSession(graph);
// TODO: Make this a loop over a dynamic set of graph inputs
if ((graphScope.Length > 1) && (graphScope[graphScope.Length - 1] != '/'))
{
graphScope = graphScope + '/';
}
if (graph[graphScope + BatchSizePlaceholderName] != null)
{
hasBatchSize = true;
}
if ((graph[graphScope + RecurrentInPlaceholderName] != null) &&
(graph[graphScope + RecurrentOutPlaceholderName] != null))
{
hasRecurrent = true;
var runner = session.GetRunner();
runner.Fetch(graph[graphScope + "memory_size"][0]);
var networkOutput = runner.Run()[0].GetValue();
memorySize = (int)networkOutput;
}
if (graph[graphScope + VectorObservationPlacholderName] != null)
{
hasState = true;
}
if (graph[graphScope + PreviousActionPlaceholderName] != null)
{
hasPrevAction = true;
}
if (graph[graphScope + "value_estimate"] != null)
{
hasValueEstimate = true;
}
if (graph[graphScope + ActionMaskPlaceholderName] != null)
{
hasMaskedActions = true;
}
}
observationMatrixList = new List <float[, , , ]>();
texturesHolder = new List <Texture2D>();
#endif
}
/// <summary>
/// Initializes the environment, configures it and initialized the Academy.
/// </summary>
private void InitializeEnvironment()
{
// Retrieve Brain and initialize Academy
var brains = GetBrains(gameObject);
InitializeAcademy();
Communicator communicator = null;
// Try to launch the communicator by usig the arguments passed at launch
try
{
communicator = new RPCCommunicator(
new CommunicatorParameters
{
port = ReadArgs()
});
}
// If it fails, we check if there are any external brains in the scene
// If there are : Launch the communicator on the default port
// If there arn't, there is no need for a communicator and it is set
// to null
catch
{
communicator = null;
var externalBrain = brains.FirstOrDefault(b => b.brainType == BrainType.External);
if (externalBrain != null)
{
communicator = new RPCCommunicator(
new CommunicatorParameters
{
port = 5005
});
}
}
brainBatcher = new Batcher(communicator);
// Initialize Brains and communicator (if present)
foreach (var brain in brains)
{
brain.InitializeBrain(this, brainBatcher);
}
if (communicator != null)
{
isCommunicatorOn = true;
var academyParameters =
new UnityRLInitializationOutput();
academyParameters.Name = gameObject.name;
academyParameters.Version = kApiVersion;
foreach (var brain in brains)
{
var bp = brain.brainParameters;
academyParameters.BrainParameters.Add(
Batcher.BrainParametersConvertor(
bp,
brain.gameObject.name,
(BrainTypeProto)
brain.brainType));
}
academyParameters.EnvironmentParameters =
new EnvironmentParametersProto();
foreach (var key in resetParameters.Keys)
{
academyParameters.EnvironmentParameters.FloatParameters.Add(
key, resetParameters[key]
);
}
var pythonParameters = brainBatcher.SendAcademyParameters(academyParameters);
Random.InitState(pythonParameters.Seed);
Application.logMessageReceived += HandleLog;
logPath = Path.GetFullPath(".") + "/unity-environment.log";
logWriter = new StreamWriter(logPath, false);
logWriter.WriteLine(DateTime.Now.ToString());
logWriter.WriteLine(" ");
logWriter.Close();
}
// If a communicator is enabled/provided, then we assume we are in
// training mode. In the absence of a communicator, we assume we are
// in inference mode.
isInference = !isCommunicatorOn;
BrainDecideAction += () => { };
AgentSetStatus += (m, d, i) => { };
AgentResetIfDone += () => { };
AgentSendState += () => { };
AgentAct += () => { };
AgentForceReset += () => { };
// Configure the environment using the configurations provided by
// the developer in the Editor.
SetIsInference(!brainBatcher.GetIsTraining());
ConfigureEnvironment();
}