This has been tuned with countless runs against other bots. So pointless moves need to be removed from the search tree. By considering the pointless chase, it pushes the player's successful attack past the horizon of the search tree. But often the opponent has a pointless chase elsewhere on the board. Thus the attack is seen to be successful in only 8 ply (not 9, because quiescent search awards the capture at 8 ply). Then Red Seven approaches again and the sequence repeats. This is because after Red Seven moves up, Blue Eight moves left, Red Seven moves left, it is pointless for Blue Eight to move back, because Blue began the Two Squares sequence moving back cannot change the outcome. But with pointless move reduction it only takes a maximum of 8 ply. Without pointless move reduction, it would take many ply to determine that Red Seven can successfully win Blue Eight. For example, a common attack involves a win based on the defender limited out by the Two Squares rule, such as in the example below: Pointless moves reduce the effective depth of the search. Eliminate pointless moves from the search tree. Restrict move generation to only those pieces that can affect the outcome. If the Two Squares Settings box is checked, making the opponent abide by the Two Squares rule, then chase attacks would be much more successful and worthy of deep analysis. Can similar code be used to determine attack plans as well? The AI always abides by the Two Squares rule. There is code that forward prunes the search tree for defending against chases. Weight suspected ranks until confirmed through attack byįorward pruning will be required to get to deeper levels. However, even if the opponent does bluff, the AI does not heavily Which allows the AI to easily determine piece ranks The result is a stratego bot with a modest amateur level of play,Īble to defeat most casual human players and stratego bots.Ī primary failing of stratego bots is the lack of bluffing, Which results in cautious play until the suspected ranksĪre confirmed through attack or mature through extended play. The AI is conservative in the assignment of suspected rank, So the generic minimax algorithm is able to determine the optimal result. The AI relies on the minimax algorithm to findīecause every unknown piece is assigned a suspected rank, Once a piece arrives in the general vicinity of its target(s), Which are mitigated only by the Two-Squares and More-Squares rules. This does lead the AI to make pointless chases However, because the outcome cannot be determined at this point, Where material or informational gain could be possible. To move its pieces towards opponent pieces The AI creates plans (using a maze-running algorithm) Success is largely dependent on the AI in gaining Static position analysis (aka oracle or pre-processing) Or protector, the AI tries to attack it with pieces ofĮxpendable rank to confirm its suspicions. When an opponent piece appears to act as a strong chase piece Opponent piece interacts with its pieces. The AI derives opponent piece rank based on how an Which allow it attack all higher ranked or unknown moved pieces with impunity. The AI's low ranked pieces (Two and up) become locally invincible, While it tries to discover the ranks of the opponent's low ranked pieces,Īs the opponent's low ranked pieces become known, Leads to key opponent piece discovery that assists in winning the game.Ī primary goal of the AI is to keep its low ranked pieces unknown Whereas lesser AI pieces are eager to sacrifice themselves if the loss Hence, an valuable AI piece will generally avoid unknown and unmoved opponent pieces Of discovering important piece ranks versus the risk of loss. It weighs which pieces to attack based on the potential The evaluation function is based on material gained versus lostĪs well as opponent piece rank discovery. Usually completing about 6 ply depending on the speed of the desktop. The search time is adjustable with the Settings menu. Move generation abides by the Two-Square and More-Squares rules. The horizon effect resulting from a very shallow search depth. Quiescent Search and Extended Search are used to reduce Iterative deepening, transposition table, heuristic history and killer move are usedįor move ordering to increase pruning success. The AI algorithm is the generic minimax with alpha-beta pruning. For a more challenging game, click the Settings menu and set the difficulty level to the middle (about 1 second).įor an easier game, keep the difficulty all the way to the leftĪnd disable the Two Squares rule.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |