public override QState Run(QState currentState, int trialNum, decimal learn, decimal discount, decimal explore)
{
QSearch qsearch = new QSearch(this);
QSearchResult actions = qsearch.AStar(currentState, true);
if (actions != null)
{
foreach (QAction action in actions.actionsList)
{
if (!currentState.IsEnd() && isRunning && currentState.GetActions().Contains(action))
{
WriteOutput(currentState + ": " + action);
QState newState = currentState.GetNewState(action);
newState.Inherit(currentState);
newState.Step();
currentState = newState;
}
}
if (currentState.IsEnd()) WriteOutput(currentState + ": End");
else
{
WriteOutput("Existing solution no longer applicable. Re-solving...");
return Run(currentState, trialNum, learn, discount, explore);
}
}
else WriteOutput("No solution found.", true);
return currentState;
}
public override Dictionary<QFeature, decimal> GetFeatures(QAction action)
{
Point self = ((Maze_With_No_Walls)GetNewState(action)).self;
QSearch qsearch = new QSearch(this);
Maze_With_No_Walls simpleMaze = new Maze_With_No_Walls() { maze = maze, self = self, goal = goal, width = width, height = height };
QSearchResult bestPath = qsearch.AStar(simpleMaze);
return QFeature_String.FromStringDecimalDictionary(new Dictionary<string, decimal>() {
//{this.GetHashCode().ToString()+"_"+action, 1}, // Identity to convert this back to QLearning
{"Goal", goal==self? 1:0},
{"Distance_To_Goal", bestPath==null? 1:(decimal) bestPath.Count / (decimal)(width * height)},
});
}
public override Dictionary<QFeature, decimal> GetFeatures(QAction action)
{
Point self = ((Maze)GetNewState(action)).self;
QSearch qsearch = new QSearch(this);
Maze simpleMaze = new Maze() { maze = maze, self = self, goal = goal, width = width, height = height, walls = walls.ToArray() };
QSearchResult bestPath = qsearch.AStar(simpleMaze);
List<Point> bestOppMoves = new List<Point>();
foreach (Point o in opponent)
{
bestOppMoves.Add(o);
bestOppMoves.Add(new Point(o.X + 1, o.Y));
bestOppMoves.Add(new Point(o.X - 1, o.Y));
bestOppMoves.Add(new Point(o.X, o.Y + 1));
bestOppMoves.Add(new Point(o.X, o.Y - 1));
}
foreach (Point o in bestOppMoves.ToArray())
{
bestOppMoves.Add(new Point(o.X + 1, o.Y));
bestOppMoves.Add(new Point(o.X - 1, o.Y));
bestOppMoves.Add(new Point(o.X, o.Y + 1));
bestOppMoves.Add(new Point(o.X, o.Y - 1));
}
Maze safeMaze = new Maze() { maze = maze, self = self, goal = goal, width = width, height = height, walls = walls.Concat(bestOppMoves).ToArray() };
QSearchResult safePath = qsearch.AStar(safeMaze);
Dictionary<string, decimal> features = new Dictionary<string, decimal>() {
//{this.GetHashCode().ToString()+"_"+action, 1}, // Identity to convert this back to QLearning
{"Goal", goal==self? 1:0},
{"Direct_Distance_To_Goal", bestPath==null? 1:(decimal) bestPath.Count / (decimal)(width * height)},
{"Safe_Distance_To_Goal", (safePath==null? 1: (decimal)safePath.Count/ (decimal)(width * height))}
};
decimal distanceToOpponent = decimal.MaxValue;
if (opponent.Any())
{
features["Opponent"] = goal!=self && opponent.Where(p => (Math.Abs(p.X - self.X) <= 1 && p.Y == self.Y) || (Math.Abs(p.Y - self.Y) <= 1 && p.X == self.X)).Any() ? 1 : 0;
distanceToOpponent = opponent.Select(o => qsearch.AStar(new Maze() { maze = maze, self = self, goal = o, width = width, height = height, walls = walls })).Select(x=>x==null? width*height:x.Count).Min();
features["Distance_To_Opponent"] = distanceToOpponent>=5? 1:distanceToOpponent / (decimal)(width * height);
if (goal != self)
{
Maze deadEnd = new Maze() { maze = maze, self = self, goal = goal, width = width, height = height, walls = walls.Concat(new Point[] { new Point(this.self.X - 1, this.self.Y), new Point(this.self.X + 1, this.self.Y), new Point(this.self.X, this.self.Y - 1), new Point(this.self.X, this.self.Y + 1) }).ToArray() };
QSearchResult deadPath = qsearch.Depth_First(deadEnd);
if (deadPath == null)
{
features["Dead_End"] = 1;
}
}
}
return QFeature_String.FromStringDecimalDictionary(features);
}
