/// <summary>
///
/// </summary>
/// <param name="actor"></param>
/// <param name="critic"></param>
/// <param name="actionsNum"></param>
/// <param name="featuresNum"></param>
/// <param name="actionKeys"></param>
public DQN(
ISupportNet actor, ISupportNet critic,
int actionsNum, int featuresNum,
int[] actionKeys)
{
_actionsNumber = actionsNum;
_featuresNumber = featuresNum;
_actionKeys = actionKeys;
_actorNet = actor;
_criticNet = critic;
//统一学习批次
_batchSize = (int)actor.BatchSize;
}
/// <summary>
/// DQN model.
/// you should set a new env object to DQN.Env before train it.
/// and there is no need env if you apply it only
/// string modelDirectoryname, string deviceName, IEnv env, int epochs = 3000, int switchEpoch = -1
/// </summary>
/// <param name="modelFilename"></param>
/// <param name="env"></param>
/// <param name="epochs"></param>
/// <returns></returns>
public static DQN Load(
byte[] actorBuffer, byte[] criticBuffer,
int actionsNum, int featuresNum,
int[] actionKeys,
string innerTypeName,
string deviceName)
{
ISupportNet actor = null, critic = null;
//if (innerTypeName == typeof(FCN7).Name)
//{
// actor = FCN7.Load(actorBuffer, deviceName);
// critic = FCN7.Load(criticBuffer, deviceName);
//}
return(new DQN(actor, critic, actionsNum, featuresNum, actionKeys));
}
public void Accept(ISupportNet sourceNet)
{
//convert to bytes
byte[] bytes = sourceNet.PersistenceMemory();
//read model and set parameters
classifierOutput = Function.Load(bytes, device);
inputVariable = classifierOutput.Inputs.First(v => v.Name == "inputVariable");
outputVariable = Variable.InputVariable(classifierOutput.Output.Shape, DataType.Float, "labelVariable");
var trainingLoss = CNTKLib.SquaredError(classifierOutput, outputVariable);
var prediction = CNTKLib.SquaredError(classifierOutput, outputVariable);
TrainingParameterScheduleDouble learningRatePerSample = new TrainingParameterScheduleDouble(0.00178125, 1); //0.00178125
TrainingParameterScheduleDouble momentumTimeConstant = CNTKLib.MomentumAsTimeConstantSchedule(256);
IList <Learner> parameterLearners = new List <Learner>()
{
Learner.MomentumSGDLearner(classifierOutput.Parameters(), learningRatePerSample, momentumTimeConstant, true)
};
trainer = Trainer.CreateTrainer(classifierOutput, trainingLoss, prediction, parameterLearners);
}
