/// <summary>
/// Executes random actions with the next pieces over the state of node
/// </summary>
/// <param name="node"></param>
/// <returns></returns>
protected virtual float Rollout(MCTSNode node)
{
TetrisState newState = node.state.CloneState();
int nPieces = 0;
float totalScore = node.state.GetScore();
float weight = 1f;
float totalWeight = weight;
//node.height identifies the height of the node in the MCTreeSearch, but also identifies the index of the piece inside the history of all the pieces played
//So, if that node.height plus the number of pieces played in the rollout are bigger than the number of known pieces, then the rollout must stop.
//Also it stops if an action has caused a game over
while ((node.height + nPieces) < pieces.Count && !newState.IsTerminal())
{
weight *= rolloutScoreWeightReduction;
totalWeight += weight;
PieceModel piece;
piece = new PieceModel(pieces[node.height + nPieces]);
newState.DoAction(piece, newState.GetRandomAction(piece));
nPieces++;
totalScore += newState.GetScore() * weight;
}
float score = totalScore / totalWeight;
rollouts++;
return(score);
}
/// <summary>
/// Main bot method that searches for the best action to do with the current piece in the current state.
/// It has a budget of time which is the max time it has to find that best action
/// </summary>
/// <param name="nextPieceType"></param>
/// <param name="budget"></param>
/// <returns></returns>
public virtual IEnumerator ActCoroutine(PieceType nextPieceType, float budget)
{
float t0 = 0.0f;
PieceModel nextPiece = new PieceModel(nextPieceType);
List <PieceAction> possibleActions; //First of all, it gets the possible actions with the current piece in the current state
if (!pieceActionDictionary.ContainsKey(nextPieceType))
{
possibleActions = emptyTetrisState.GetActions(nextPiece);
pieceActionDictionary.Add(nextPieceType, possibleActions);
}
else
{
possibleActions = pieceActionDictionary[nextPieceType];
}
float bestScore = -float.MaxValue;
PieceAction bestAction;
int i = Random.Range(0, possibleActions.Count); //The first possible action to test is chosen randomly
int initialIndex = i;
bestAction = possibleActions[i];
yield return(null);
t0 += Time.deltaTime;
//In a while loop that goes until the time ends
while (t0 < budget && i < possibleActions.Count)
{
if (!TBController.pausedGame)
{
t0 += Time.deltaTime;
TetrisState newState = currentTetrisState.CloneState(); //The TetrisState is cloned
newState.DoAction(nextPiece, possibleActions[i]); //One of the possible actions is played in the cloned state
nextPiece.ResetCoordinates();
float score = newState.GetScore(); //And its score is got
if (score > bestScore)
{
bestScore = score;
bestAction = possibleActions[i];
}
i++;
if (i == possibleActions.Count)
{
i = 0;
}
if (i == initialIndex)
{
break; //If all the possible actions have been tested, the while loop ends
}
}
yield return(null);
}
currentTetrisState.DoAction(nextPiece, bestAction); //The bestAction is played in the real state
TBController.DoActionByBot(bestAction); //And also, it is said to the TetrisBoardController to play that action in the real board
}
