/// <summary>
/// Sets the Controller that the bot represents within a SabberStone Game.
/// </summary>
/// <param name="controller">This bot's Controller.</param>
public void SetController(Controller controller)
{
Player = controller;
RandomPlayoutBot = new RandomBot(controller, filterDuplicatePositionTasks: true);
// Also reset the table of statistics
ResetMASTData();
}
/// <summary>
/// Creates a new instance.
/// </summary>
/// <param name="playout">The strategy used to play out a game in simulation.</param>
/// <param name="evaluation">The evaluation strategy for determining the value of samples.</param>
/// <param name="gameLogic">The game specific logic required for searching through SabberStoneStates and SabberStoneActions.</param>
public SabberStoneSideInformationStrategy(IPlayoutStrategy <List <SabberStoneAction>, SabberStoneState, SabberStoneAction, object, SabberStoneAction> playout, IStateEvaluation <List <SabberStoneAction>, SabberStoneState, SabberStoneAction, object, SabberStoneAction, TreeSearchNode <SabberStoneState, SabberStoneAction> > evaluation, IGameLogic <List <SabberStoneAction>, SabberStoneState, SabberStoneAction, object, SabberStoneAction, SabberStoneAction> gameLogic)
{
PlayoutBot = new RandomBot();
Playout = playout;
Evaluation = evaluation;
GameLogic = gameLogic;
}
/// <summary>
/// Creates a new instance of the MASTPlayoutBot.
/// </summary>
/// <param name="selection">The type of selection to use.</param>
/// <param name="evaluation">The EvaluationStrategy used for evaluating a SabberStoneState.</param>
/// <param name="eGreedyThreshold">[Optional] Threshold for e-greedy selection. Default value is <see cref="Constants.DEFAULT_E_GREEDY_THRESHOLD"/>.</param>
/// <param name="ucbConstantC">[Optional] Value for the c-constant in the UCB1 formula. Default value is <see cref="Constants.DEFAULT_UCB1_C"/>.</param>
public MASTPlayoutBot(SelectionType selection, EvaluationStrategyHearthStone evaluation, double eGreedyThreshold = Constants.DEFAULT_E_GREEDY_THRESHOLD, double ucbConstantC = Constants.DEFAULT_UCB1_C)
{
Selection = selection;
RandomPlayoutBot = new RandomBot(filterDuplicatePositionTasks: true);
Evaluation = evaluation;
EGreedyThreshold = eGreedyThreshold;
UCBConstantC = ucbConstantC;
MASTTable = new Dictionary <int, PlayerTaskStatistics>();
ActionsTaken = new List <SabberStoneAction>();
}
/// <summary>
/// Constructs a new instance of LSIBot with default strategies.
/// </summary>
/// <param name="allowPerfectInformation">[Optional] Whether or not this bot is allowed perfect information about the game state (i.e. no obfuscation and therefore no determinisation). Default value is false.</param>
/// <param name="ensembleSize">[Optional] The size of the ensemble to use. Default value is 1.</param>
/// <param name="playoutBotType">[Optional] The type of playout bot to be used during playouts. Default value is <see cref="PlayoutBotType.MAST"/>.</param>
/// <param name="mastSelectionType">[Optional] The type of selection strategy used by the M.A.S.T. playout. Default value is <see cref="MASTPlayoutBot.SelectionType.EGreedy"/>.</param>
/// <param name="playoutTurnCutoff">[Optional] The amount of turns after which to stop a simulation. Default value is <see cref="Constants.DEFAULT_PLAYOUT_TURN_CUTOFF"/>.</param>
/// <param name="budgetType">[Optional] The type of budget that this bot will use. Default value is <see cref="BudgetType.Iterations"/>.</param>
/// <param name="samples">[Optional] The budget for the amount of iterations LSI can use. Default value is <see cref="Constants.DEFAULT_COMPUTATION_ITERATION_BUDGET"/>.</param>
/// <param name="time">[Optional] The budget for the amount of milliseconds LSI can spend on searching. Default value is <see cref="Constants.DEFAULT_COMPUTATION_TIME_BUDGET"/>.</param>
/// <param name="generationBudgetPercentage">[Optional] The percentage of the budget that should be spent during the generation phase. Default value is <see cref="Constants.DEFAULT_LSI_BUDGET_GENERATION_PERCENTAGE"/>.</param>
/// <param name="budgetEstimationType">[Optional] The type of strategy used to estimate the budget for LSI. Default value is <see cref="BudgetEstimationType.AverageSampleTime"/>.</param>
/// <param name="useHeuristicEvaluation">[Optional] Whether or not to use the HeuristicBot's evaluation function. Default value is false.</param>
/// <param name="debugInfoToConsole">[Optional] Whether or not to write debug information to the console. Default value is false.</param>
public LSIBot(bool allowPerfectInformation = false,
int ensembleSize = 1,
PlayoutBotType playoutBotType = PlayoutBotType.MAST,
MASTPlayoutBot.SelectionType mastSelectionType = MASTPlayoutBot.SelectionType.EGreedy,
int playoutTurnCutoff = Constants.DEFAULT_PLAYOUT_TURN_CUTOFF,
BudgetType budgetType = BudgetType.Iterations,
int samples = Constants.DEFAULT_COMPUTATION_ITERATION_BUDGET,
long time = Constants.DEFAULT_COMPUTATION_TIME_BUDGET,
double generationBudgetPercentage = Constants.DEFAULT_LSI_BUDGET_GENERATION_PERCENTAGE,
BudgetEstimationType budgetEstimationType = BudgetEstimationType.AverageSampleTime,
bool useHeuristicEvaluation = false,
bool debugInfoToConsole = false)
{
PerfectInformation = allowPerfectInformation;
EnsembleSize = ensembleSize;
PlayoutBotType = playoutBotType;
MASTSelectionType = mastSelectionType;
PlayoutTurnCutoff = playoutTurnCutoff;
BudgetType = budgetType;
Samples = samples;
Time = time;
GenerationBudgetPercentage = generationBudgetPercentage;
BudgetEstimation = budgetEstimationType;
_debug = debugInfoToConsole;
// Create the ensemble search
Ensemble = new EnsembleStrategySabberStone(enableStateObfuscation: true, enablePerfectInformation: PerfectInformation);
// Adjust sample sizes for use in the Ensemble
long budgetAllowance;
switch (BudgetType)
{
case BudgetType.Iterations:
Samples = EnsembleSize > 0 ? Samples / EnsembleSize : Samples; // Note: Integer division by design.
budgetAllowance = Samples;
break;
case BudgetType.Time:
Time = EnsembleSize > 0 ? Time / EnsembleSize : Time; // Note: Integer division by design.
budgetAllowance = Time;
break;
default:
throw new InvalidEnumArgumentException($"BudgetType `{BudgetType}' is not supported.");
}
// Simulation will be handled by the Playout.
var sabberStoneStateEvaluation = new EvaluationStrategyHearthStone(useHeuristicEvaluation);
Playout = new PlayoutStrategySabberStone();
// Set the playout bots
switch (PlayoutBotType)
{
case PlayoutBotType.Random:
MyPlayoutBot = new RandomBot(filterDuplicatePositionTasks: true);
OpponentPlayoutBot = new RandomBot(filterDuplicatePositionTasks: true);
break;
case PlayoutBotType.Heuristic:
MyPlayoutBot = new HeuristicBot();
OpponentPlayoutBot = new HeuristicBot();
break;
case PlayoutBotType.MAST:
MyPlayoutBot = new MASTPlayoutBot(MASTSelectionType, sabberStoneStateEvaluation);
OpponentPlayoutBot = new MASTPlayoutBot(MASTSelectionType, sabberStoneStateEvaluation);
break;
default:
throw new InvalidEnumArgumentException($"PlayoutBotType `{PlayoutBotType}' is not supported.");
}
// Set the budget estimation strategy.
switch (BudgetEstimation)
{
case BudgetEstimationType.AverageSampleTime:
BudgetEstimationStrategy = new AverageSampleTimeBudgetEstimationStrategy(BudgetType, budgetAllowance, GenerationBudgetPercentage, Playout);
break;
case BudgetEstimationType.PreviousSearchAverage:
BudgetEstimationStrategy = new PreviousSearchAverageBudgetEstimationStrategy(BudgetType, budgetAllowance, GenerationBudgetPercentage, Playout);
break;
default:
throw new InvalidEnumArgumentException($"BudgetEstimationType `{budgetEstimationType}' is not supported.");
}
// LSI will need a goal-strategy to determine when a simulation is done
Goal = new GoalStrategyTurnCutoff(PlayoutTurnCutoff);
// LSI will need an evaluation-strategy to evaluate the strength of samples
Evaluation = new EvaluationStrategyHearthStone();
// Application will be handled by the GameLogic
// Hierarchical Expansion is set to TRUE because of incremental task validation during action sampling.
GameLogic = new SabberStoneGameLogic(Goal, hierarchicalExpansion: true);
// The side information strategy needs access to several of these.
SideInformationStrategy = new SabberStoneSideInformationStrategy(Playout, Evaluation, GameLogic);
// The sampling strategy used to sample actions during the generation phase.
SamplingStrategy = new SabberStoneLSISamplingStrategy(GameLogic);
}
/// <summary>
/// Constructs a new instance of MCTSBot with default strategies.
/// </summary>
/// <param name="allowPerfectInformation">[Optional] Whether or not this bot is allowed perfect information about the game state (i.e. no obfuscation and therefore no determinisation). Default value is false.</param>
/// <param name="ensembleSize">[Optional] The size of the ensemble to use. Default value is 1.</param>
/// <param name="playoutBotType">[Optional] The type of playout bot to be used during playouts. Default value is <see cref="PlayoutBotType.MAST"/>.</param>
/// <param name="mastSelectionType">[Optional] The type of selection strategy used by the MAST playout. Default value is <see cref="MASTPlayoutBot.SelectionType.EGreedy"/>.</param>
/// <param name="retainTaskStatistics">[Optional] Whether or not to retain the PlayerTask statistics between searches. Default value is false.</param>
/// <param name="budgetType">[Optional] The type of budget that this bot will use. Default value is <see cref="BudgetType.Iterations"/>.</param>
/// <param name="iterations">[Optional] The budget for the amount of iterations MCTS can use. Default value is <see cref="Constants.DEFAULT_COMPUTATION_ITERATION_BUDGET"/>.</param>
/// <param name="time">[Optional] The budget for the amount of milliseconds MCTS can spend on searching. Default value is <see cref="Constants.DEFAULT_COMPUTATION_TIME_BUDGET"/>.</param>
/// <param name="minimumVisitThresholdForExpansion">[Optional] The minimum amount of times a node has to be visited before it can be expanded. Default value is <see cref="Constants.DEFAULT_MCTS_MINIMUM_VISIT_THRESHOLD_FOR_EXPANSION"/>.</param>
/// <param name="minimumVisitThresholdForSelection">[Optional] The minimum number of visits before using the NodeEvaluation to select the best node. Default value is <see cref="Constants.DEFAULT_MCTS_MINIMUM_VISIT_THRESHOLD_FOR_SELECTION"/>.</param>
/// <param name="playoutTurnCutoff">[Optional] The amount of turns after which to stop a simulation. Default value is <see cref="Constants.DEFAULT_PLAYOUT_TURN_CUTOFF"/>.</param>
/// <param name="ucbConstantC">[Optional] Value for the c-constant in the UCB1 formula. Default value is <see cref="Constants.DEFAULT_UCB1_C"/>.</param>
/// <param name="dimensionalOrdering">[Optional] The ordering for dimensions when using Hierarchical Expansion. Default value is <see cref="DimensionalOrderingType.None"/>.</param>
/// <param name="useHeuristicEvaluation">[Optional] Whether or not to use the HeuristicBot's evaluation function. Default value is false.</param>
/// <param name="debugInfoToConsole">[Optional] Whether or not to write debug information to the console. Default value is false.</param>
public HMCTSBot(bool allowPerfectInformation = false,
int ensembleSize = 1,
PlayoutBotType playoutBotType = PlayoutBotType.MAST,
MASTPlayoutBot.SelectionType mastSelectionType = MASTPlayoutBot.SelectionType.EGreedy,
bool retainTaskStatistics = false,
BudgetType budgetType = BudgetType.Iterations,
int iterations = Constants.DEFAULT_COMPUTATION_ITERATION_BUDGET,
long time = Constants.DEFAULT_COMPUTATION_TIME_BUDGET,
int minimumVisitThresholdForExpansion = Constants.DEFAULT_MCTS_MINIMUM_VISIT_THRESHOLD_FOR_EXPANSION,
int minimumVisitThresholdForSelection = Constants.DEFAULT_MCTS_MINIMUM_VISIT_THRESHOLD_FOR_SELECTION,
int playoutTurnCutoff = Constants.DEFAULT_PLAYOUT_TURN_CUTOFF,
double ucbConstantC = Constants.DEFAULT_UCB1_C,
DimensionalOrderingType dimensionalOrdering = DimensionalOrderingType.None,
bool useHeuristicEvaluation = false,
bool debugInfoToConsole = false)
{
PerfectInformation = allowPerfectInformation;
EnsembleSize = ensembleSize;
PlayoutBotType = playoutBotType;
MASTSelectionType = mastSelectionType;
RetainTaskStatistics = retainTaskStatistics;
BudgetType = budgetType;
Iterations = iterations;
Time = time;
MinimumVisitThresholdForExpansion = minimumVisitThresholdForExpansion;
MinimumVisitThresholdForSelection = minimumVisitThresholdForSelection;
PlayoutTurnCutoff = playoutTurnCutoff;
UCBConstantC = ucbConstantC;
DimensionalOrdering = dimensionalOrdering;
_debug = debugInfoToConsole;
// Create the ensemble search
Ensemble = new EnsembleStrategySabberStone(enableStateObfuscation: true, enablePerfectInformation: PerfectInformation);
// Simulation will be handled by the Playout.
var sabberStoneStateEvaluation = new EvaluationStrategyHearthStone(useHeuristicEvaluation);
Playout = new PlayoutStrategySabberStone();
// Set the playout bots
switch (PlayoutBotType)
{
case PlayoutBotType.Random:
MyPlayoutBot = new RandomBot(filterDuplicatePositionTasks: true);
OpponentPlayoutBot = new RandomBot(filterDuplicatePositionTasks: true);
break;
case PlayoutBotType.Heuristic:
MyPlayoutBot = new HeuristicBot();
OpponentPlayoutBot = new HeuristicBot();
break;
case PlayoutBotType.MAST:
MyPlayoutBot = new MASTPlayoutBot(MASTSelectionType, sabberStoneStateEvaluation);
OpponentPlayoutBot = new MASTPlayoutBot(MASTSelectionType, sabberStoneStateEvaluation);
break;
default:
throw new InvalidEnumArgumentException($"PlayoutBotType `{PlayoutBotType}' is not supported.");
}
// We'll be cutting off the simulations after X turns, using a GoalStrategy.
Goal = new GoalStrategyTurnCutoff(PlayoutTurnCutoff);
// Expansion, Application and Goal will be handled by the GameLogic.
GameLogic = new SabberStoneGameLogic(Goal, true, DimensionalOrdering);
// Create the INodeEvaluation strategy used in the selection phase.
var nodeEvaluation = new ScoreUCB <SabberStoneState, SabberStoneAction>(UCBConstantC);
// Build MCTS
Builder = MCTS <List <SabberStoneAction>, SabberStoneState, SabberStoneAction, object, SabberStoneAction> .Builder();
Builder.ExpansionStrategy = new MinimumTExpansion <List <SabberStoneAction>, SabberStoneState, SabberStoneAction, object, SabberStoneAction>(MinimumVisitThresholdForExpansion);
Builder.SelectionStrategy = new BestNodeSelection <List <SabberStoneAction>, SabberStoneState, SabberStoneAction, object, SabberStoneAction>(MinimumVisitThresholdForSelection, nodeEvaluation);
Builder.EvaluationStrategy = sabberStoneStateEvaluation;
switch (BudgetType)
{
case BudgetType.Iterations:
Builder.Iterations = EnsembleSize > 0 ? Iterations / EnsembleSize : Iterations; // Note: Integer division by design.
break;
case BudgetType.Time:
Builder.Time = EnsembleSize > 0 ? Time / EnsembleSize : Time; // Note: Integer division by design.
break;
default:
throw new InvalidEnumArgumentException($"BudgetType `{BudgetType}' is not supported.");
}
Builder.BackPropagationStrategy = new EvaluateOnceAndColourBackPropagation <List <SabberStoneAction>, SabberStoneState, SabberStoneAction, object, SabberStoneAction>();
Builder.FinalNodeSelectionStrategy = new BestRatioFinalNodeSelection <List <SabberStoneAction>, SabberStoneState, SabberStoneAction, object, SabberStoneAction>();
Builder.SolutionStrategy = new SolutionStrategySabberStone(true, nodeEvaluation);
Builder.PlayoutStrategy = Playout;
}
/// <summary>
/// Constructs a new instance.
/// </summary>
/// <param name="naïveBot">The bot to use when creating samples.</param>
public SabberStoneNMCSamplingStrategy(RandomBot naïveBot)
{
NaïveBot = naïveBot;
}
/// <summary>
/// Constructs a new instance of NMCTSBot with default strategies.
/// </summary>
/// <param name="allowPerfectInformation">[Optional] Whether or not this bot is allowed perfect information about the game state (i.e. no obfuscation and therefore no determinisation). Default value is false.</param>
/// <param name="ensembleSize">[Optional] The size of the ensemble to use. Default value is 1.</param>
/// <param name="playoutBotType">[Optional] The type of playout bot to be used during playouts. Default value is <see cref="PlayoutBotType.MAST"/>.</param>
/// <param name="mastSelectionType">[Optional] The type of selection strategy used by the MAST playout. Default value is <see cref="MASTPlayoutBot.SelectionType.EGreedy"/>.</param>
/// <param name="budgetType">[Optional] The type of budget that this bot will use. Default value is <see cref="BudgetType.Iterations"/>.</param>
/// <param name="iterations">[Optional] The budget for the amount of iterations NMCTS can use. Default value is <see cref="Constants.DEFAULT_COMPUTATION_ITERATION_BUDGET"/>.</param>
/// <param name="time">[Optional] The budget for the amount of milliseconds NMCTS can spend on searching. Default value is <see cref="Constants.DEFAULT_COMPUTATION_TIME_BUDGET"/>.</param>
/// <param name="playoutTurnCutoff">[Optional] The amount of turns after which to stop a simulation. Default value is <see cref="Constants.DEFAULT_PLAYOUT_TURN_CUTOFF"/>.</param>
/// <param name="globalPolicy">[Optional] The exploration-threshold for the e-greedy global policy. Default value is <see cref="Constants.DEFAULT_NMCTS_GLOBAL_POLICY"/>.</param>
/// <param name="localPolicy">[Optional] The exploration-threshold for the e-greedy local policy. Default value is <see cref="Constants.DEFAULT_NMCTS_LOCAL_POLICY"/>.</param>
/// <param name="useHeuristicEvaluation">[Optional] Whether or not to use the HeuristicBot's evaluation function. Default value is false.</param>
/// <param name="debugInfoToConsole">[Optional] Whether or not to write debug information to the console. Default value is false.</param>
public NMCTSBot(bool allowPerfectInformation = false,
int ensembleSize = 1,
PlayoutBotType playoutBotType = PlayoutBotType.MAST,
MASTPlayoutBot.SelectionType mastSelectionType = MASTPlayoutBot.SelectionType.EGreedy,
BudgetType budgetType = BudgetType.Iterations,
int iterations = Constants.DEFAULT_COMPUTATION_ITERATION_BUDGET,
long time = Constants.DEFAULT_COMPUTATION_TIME_BUDGET,
int playoutTurnCutoff = Constants.DEFAULT_PLAYOUT_TURN_CUTOFF,
double globalPolicy = Constants.DEFAULT_NMCTS_GLOBAL_POLICY,
double localPolicy = Constants.DEFAULT_NMCTS_LOCAL_POLICY,
bool useHeuristicEvaluation = false,
bool debugInfoToConsole = false)
{
PerfectInformation = allowPerfectInformation;
EnsembleSize = ensembleSize;
PlayoutBotType = playoutBotType;
MASTSelectionType = mastSelectionType;
BudgetType = budgetType;
Iterations = iterations;
Time = time;
PlayoutTurnCutoff = playoutTurnCutoff;
GlobalPolicy = globalPolicy;
LocalPolicy = localPolicy;
_debug = debugInfoToConsole;
// Create the ensemble search
Ensemble = new EnsembleStrategySabberStone(enableStateObfuscation: true, enablePerfectInformation: PerfectInformation);
// Simulation will be handled by the Playout
var sabberStoneStateEvaluation = new EvaluationStrategyHearthStone(useHeuristicEvaluation);
Playout = new PlayoutStrategySabberStone();
// Set the playout bots
switch (PlayoutBotType)
{
case PlayoutBotType.Random:
MyPlayoutBot = new RandomBot(filterDuplicatePositionTasks: true);
OpponentPlayoutBot = new RandomBot(filterDuplicatePositionTasks: true);
break;
case PlayoutBotType.Heuristic:
MyPlayoutBot = new HeuristicBot();
OpponentPlayoutBot = new HeuristicBot();
break;
case PlayoutBotType.MAST:
MyPlayoutBot = new MASTPlayoutBot(MASTSelectionType, sabberStoneStateEvaluation);
OpponentPlayoutBot = new MASTPlayoutBot(MASTSelectionType, sabberStoneStateEvaluation);
break;
default:
throw new InvalidEnumArgumentException($"PlayoutBotType `{PlayoutBotType}' is not supported.");
}
// And the random sampling bot
RandomSamplingBot = new RandomBot(filterDuplicatePositionTasks: true);
// We'll be cutting off the simulations after X turns, using a GoalStrategy
Goal = new GoalStrategyTurnCutoff(PlayoutTurnCutoff);
// Application and Goal will be handled by the GameLogic
GameLogic = new SabberStoneGameLogic(Goal, false);
// Build NMCTS
Builder = NMCTS <List <SabberStoneAction>, SabberStoneState, SabberStoneAction, object, SabberStoneAction> .Builder();
Builder.ExplorationStrategy = new ChanceExploration <List <SabberStoneAction>, SabberStoneState, SabberStoneAction, object, SabberStoneAction>(LocalPolicy);
Builder.PlayoutStrategy = Playout;
Builder.PolicyGlobal = GlobalPolicy;
Builder.SamplingStrategy = new SabberStoneNMCSamplingStrategy(RandomSamplingBot);
Builder.SolutionStrategy = new SolutionStrategySabberStone(false, new AverageScore <SabberStoneState, SabberStoneAction>());
Builder.EvaluationStrategy = sabberStoneStateEvaluation;
switch (BudgetType)
{
case BudgetType.Iterations:
Builder.Iterations = EnsembleSize > 0 ? Iterations / EnsembleSize : Iterations; // Note: Integer division by design.
break;
case BudgetType.Time:
Builder.Time = EnsembleSize > 0 ? Time / EnsembleSize : Time; // Note: Integer division by design.
break;
default:
throw new InvalidEnumArgumentException($"BudgetType `{BudgetType}' is not supported.");
}
}