private const double GAMMA = 0.99; // The value of gamma to use in the Q-Learning rule
public QLearning(GridWorld gridWorld, Cell startingCell, Cell rewardCell)
{
this.currentCell = startingCell;
this.rewardCell = rewardCell;
this.gridWorld = gridWorld;
this.qTable = new double[gridWorld.GetCells().GetLength(0), gridWorld.GetCells().GetLength(1), NUMBER_OF_MOVES];
}
private bool training = true; // boolean flag to indicate training versus testing
#endregion Fields
#region Constructors
public QLearning(GridWorld gridWorld, Cell startingCell, Cell rewardCell)
{
this.currentCell = startingCell;
this.rewardCell = rewardCell;
this.gridWorld = gridWorld;
this.qTable = new double[gridWorld.GetCells().GetLength(0),gridWorld.GetCells().GetLength(1),NUMBER_OF_MOVES];
}
/// <summary>
/// Runs a tick of the A* search
/// </summary>
/// <param name="sender"></param>
/// <param name="e"></param>
public void Run(object sender, EventArgs e)
{
System.Console.Out.WriteLine("Running A*; Step Count: " + stepCount++);
while (true)
{
if (unvisitedCells.Count == 0)
{
return;
}
currentCell = unvisitedCells.Last.Value;
unvisitedCells.RemoveLast();
if (currentCell.HasBeenSearched())
{
continue;
}
break;
}
currentCell.SetHasBeenSearched(true);
visitedCells.AddLast(currentCell);
if (currentCell.IsRewardCell())
{
System.Console.Out.WriteLine("Found the reward");
return;
}
// Check if cell is a neighbor of the currentCell
// If so, set the cost and add to unvisitedCells in ProcessNeighbor()
if (gridWorld.CanMove(Direction.UP, currentCell.GetRowIndex(), currentCell.GetColumnIndex()))
{
ProcessNeighbor(gridWorld.GetCells()[currentCell.GetRowIndex() - 1, currentCell.GetColumnIndex()]);
}
if (gridWorld.CanMove(Direction.DOWN, currentCell.GetRowIndex(), currentCell.GetColumnIndex()))
{
ProcessNeighbor(gridWorld.GetCells()[currentCell.GetRowIndex() + 1, currentCell.GetColumnIndex()]);
}
if (gridWorld.CanMove(Direction.LEFT, currentCell.GetRowIndex(), currentCell.GetColumnIndex()))
{
ProcessNeighbor(gridWorld.GetCells()[currentCell.GetRowIndex(), currentCell.GetColumnIndex() - 1]);
}
if (gridWorld.CanMove(Direction.RIGHT, currentCell.GetRowIndex(), currentCell.GetColumnIndex()))
{
ProcessNeighbor(gridWorld.GetCells()[currentCell.GetRowIndex(), currentCell.GetColumnIndex() + 1]);
}
}
/// <summary>
/// Restarts an episode of Q-Learning and moves the current position to the provided starting position if one is provided.
/// </summary>
/// <param name="startingPosition">The starting position for the episode. If null is provided a random starting position will be determined.</param>
public void RestartEpisode(Cell startingPosition)
{
// Reset the step count for the next episode
stepCount = 0;
// Increment the episode counter
episodeCount++;
List <Cell> availableCells = new List <Cell>();
// Place the agent at a random tile (not an obstacle)
foreach (Cell cell in gridWorld.GetCells())
{
if (!cell.IsObstacle() && !cell.IsRewardCell())
{
availableCells.Add(cell);
}
}
if (startingPosition == null)
{
int nextCellIndex = (new Random()).Next(0, availableCells.Count - 1);
currentCell = availableCells[nextCellIndex];
}
else
{
currentCell = startingPosition;
}
}