/// <summary>
/// Uses ML.NET library
/// with the data of approximately 1000 games, and few thousand proven good moves to "predict" a "good" move
/// </summary>
private void MachineLearningMove()
{
// what was the tile placed previously
string tileplayed = (Game.PlayerTurn == 1) ? "X" : "O";
// Add input data
// The gameboard is a 2D array of size n
// Needs updating if support for >3 size is added in future
var input = new ModelInput();
input.Tile1 = Game.Gameboard.getboard()[0, 0].Value.ToString();
input.Tile2 = Game.Gameboard.getboard()[1, 0].Value.ToString();
input.Tile3 = Game.Gameboard.getboard()[2, 0].Value.ToString();
input.Tile4 = Game.Gameboard.getboard()[0, 1].Value.ToString();
input.Tile5 = Game.Gameboard.getboard()[1, 1].Value.ToString();
input.Tile6 = Game.Gameboard.getboard()[2, 1].Value.ToString();
input.Tile7 = Game.Gameboard.getboard()[0, 2].Value.ToString();
input.Tile8 = Game.Gameboard.getboard()[1, 2].Value.ToString();
input.Tile9 = Game.Gameboard.getboard()[2, 2].Value.ToString();
input.TilePlayed = tileplayed;
// Load model and predict output of sample data
ModelOutput result = ConsumeModel.Predict(input);
// Get a list of legal moves (tiles that do not have a value)
List <Tile> legalMoves = new List <Tile>();
foreach (Tile tile in Game.Gameboard.getboard())
{
if (tile.Value == Tile.TileValue.NaN)
{
legalMoves.Add(tile);
}
}
// result of prediction
int ID = Convert.ToInt32(result.Prediction);
// Find the tile with same ID
Tile MoveTile = legalMoves.Find(X => X.ID == ID);
// machine predicted an ID thats not a legal move, give a random move instead...
if (MoveTile == null)
{
Random r = new Random();
MoveTile = legalMoves[r.Next(0, legalMoves.Count - 1)];
}
if (MoveTile.CheckTileState())
{
Game.MakeMove(MoveTile.X, MoveTile.Y, Game.Players[Game.PlayerTurn].PlayerTile);
}
else
{
// Somehow the tile was taken use a random move
Game.RANDOM_MOVE();
}
}