/// <summary>
/// 随机从数组中取出数据
/// </summary>
/// <typeparam name="T"></typeparam>
/// <param name="list"></param>
/// <returns></returns>
///
public static T RandomTake <T>(this List <T> list)
{
int index = NP.Random(list.Count);
T item = list[index];
return(item);
}
public static void RandomRemove <T>(this List <T> list, int capacity)
{
if (list.Count >= capacity)
{
int index = NP.Random(list.Count);
list.RemoveAt(index);
}
}
/// <summary>
///
/// </summary>
/// <returns></returns>
private (int x, int y, double[] classIndex) RandomAccessMemory()
{
//use actionNumber represent real types
int rawValueIndex = NP.Random(_randomSeedKeys);
Point p = _memory[rawValueIndex].RandomTake();
//current one-hot action
double[] classIndex = NP.ToOneHot(Array.IndexOf(_randomSeedKeys, rawValueIndex), ActionNum);
return(p.X, p.Y, classIndex);
}
/// <summary>
///
/// </summary>
/// <typeparam name="T"></typeparam>
/// <param name="list"></param>
/// <param name="bacthSize"></param>
/// <returns></returns>
public static List <T> RandomTakeBatch <T>(this List <T> list, int limitSize = 200)
{
int num = list.Count;
List <T> dist = new List <T>();
for (int i = 0; i < limitSize; i++)
{
int key = NP.Random(num);
dist.Add(list[key]);
}
return(dist);
}
/// <summary>
/// 获取当前actor下的action和reward
/// </summary>
/// <param name="step"></param>
/// <param name="state"></param>
/// <returns></returns>
public (double[] action, double q) EpsilonGreedy(int step, double[] state)
{
int totalEpochs = Convert.ToInt32(_epoches * 0.9);
var epsion = EpsilonCalcute(step, eps_total: totalEpochs);
if (NP.Random() < epsion)
{
return(_env.RandomAction(), 0);
}
else
{
var(action, q) = ChooseAction(state);
return(action, q);
}
}
/// <summary>
/// -----------------------------------------------------
/// * 0 | 1 | 2
/// * -----------------------
/// * 7 | 8 | 3
/// * -----------------------
/// * 6 | 5 | 4
/// </summary>
/// <returns></returns>
private (int x, int y, double[] actions) RandomAccessMemory()
{
//
int rawValueIndex = NP.Random(_randomSeedKeys);
Point pt = _memory[rawValueIndex].RandomTake();
//
double[] actions = new double[ActionNum];
//快速搜索x++方向点
List <Point> points = new List <Point>()
{
new Point(pt.X - 1, pt.Y - 1), //(-1,-1)
new Point(pt.X, pt.Y - 1), //(0,-1)
new Point(pt.X + 1, pt.Y - 1), //(1,-1)
new Point(pt.X + 1, pt.Y), //(1,0)
new Point(pt.X + 1, pt.Y + 1), //(1,1)
new Point(pt.X, pt.Y + 1), //(0,1)
new Point(pt.X - 1, pt.Y + 1), //(-1,1)
new Point(pt.X - 1, pt.Y), //(-1,0)
};
//search next point
for (int pointIndex = 0; pointIndex < ActionNum; pointIndex++)
{
Point p = points[pointIndex];
//if reach to the end, use original point
if (p.X >= _limit_x || p.X < 0 || p.Y >= _limit_y || p.Y < 0)
{
continue;
}
//store right action(one-hot)
if (_queryTable[p.X, p.Y] == rawValueIndex)
{
actions.CombineOneHot(NP.ToOneHot(pointIndex, ActionNum));
}
}
//
if (!_existActions.Exists(p => NP.Equal(p, actions)))
{
_existActions.Add(actions);
}
//
return(pt.X, pt.Y, actions);
}
/// <summary>
/// random数据集
/// </summary>
public double[] RandomAction()
{
int action = NP.Random(ActionNum);
return(NP.ToOneHot(action, ActionNum));
}
