美元票据拍卖

Question

这是对博弈论中的美元票据拍卖博弈的一个KOTH挑战。其中，一美元被卖给出价最高的人。出价增加5美分，输家也支付他们的出价。他们的想法是，为了减少损失，这两家公司都将竞购战升级至一美元以外。

希望你的机器人比这更聪明。

您将通过扩展net.ramenchef.dollarauction.DollarBidder类来创建一个机器人来玩这个游戏。您必须实现返回bot下一个出价的nextBid方法，考虑到另一个bot的上一个出价。如果有必要，您也可以使用newAuction方法为每次拍卖重置与对手的机器人类。

public abstract class DollarBidder {
    /**
     * Used by the runner to keep track of scores.
     */
    long score = 0;

    /**
     * (Optional) Prepare for the next auction.
     *
     * @param opponent The class of the opponent's bot.
     */
    public void newAuction(Class opponent) {}

    /**
     * Bid on the dollar. Bidding ends if the bid is
     * not enough to top the previous bid or both bids
     * exceed $100.
     *
     * @param opponentsBid How much money, in cents,
     *  that the opponent bid in the previous round. If
     *  this is the first round in the auction, it will
     *  be 0.
     * @return How much money to bid in this round, in
     *  cents.
     */
    public abstract int nextBid(int opponentsBid);
}

招标直至发生下列情况之一：

• nextBid抛出一个异常。如果发生这种情况，抛出例外情况的机器人支付他们以前的出价，而另一个机器人则免费获得美元。
• 任何一个bot支付的钱都不足以超过之前的出价。如果发生这种情况，两个机器人都支付他们的出价(输家支付他们以前的出价)，而赢家得到一美元。
• 这两个机器人出价超过100美元。如果发生这种情况，两个机器人都会支付100美元，而两个机器人都不会得到美元。

每个机器人组合举行2次拍卖。机器人是通过它们在这些拍卖会上获得的总利润来得分的。得分最高者获胜。

示例

GreedyBot

import net.ramenchef.dollarauction.DollarBidder;

public class GreedyBot extends DollarBidder {
    @Override
    public int nextBid(int opponentsBid) {
        return opponentsBid + 5;
    }
}

OnlyWinningMove

import net.ramenchef.dollarauction.DollarBidder;

public class OnlyWinningMove extends DollarBidder {
    @Override
    public int nextBid(int opponentsBid) {
        return 0;
    }
}

AnalystBot

不要将此作为具有分析思维的机器人的模板，而是使用ImprovedAnalystBot。

import net.ramenchef.dollarauction.DollarBidder;

// yes, this is a poor implementation, but I'm not
// going to waste my time perfecting it
public class AnalystBot extends DollarBidder {
    private DollarBidder enemy;

    @Override
    public void newAuction(Class opponent) {
        try {
            enemy = opponent.newInstance();
            enemy.newAuction(this.getClass());
        } catch (ReflectiveOperationException e) {
            enemy = null;
        }
    }

    @Override
    public int nextBid(int opponentsBid) {
        if (enemy == null)
            return 0;

        return enemy.nextBid(95) >= 100 ? 0 : 95;
    }
}

AnalystKiller

import net.ramenchef.dollarauction.DollarBidder;

public class AnalystKiller extends DollarBidder {
    private static int instances = 0;
    private final boolean tainted;

    public AnalystKiller() {
        this.tainted = instances++ != 0;
    }

    @Override
    public int nextBid(int opponentsBid) {
        if (tainted)
            throw new RuntimeException("A mysterious error occurred! >:)");

        return 0;
    }
}

附加规则

• 标准漏洞是禁止的。
• 破坏其他机器人是被允许的，但试图改变视野/方法的能见度将导致神秘的SecurityExceptions。一个例外是导致另一个机器人突破500毫秒的限制。
• 除非扩展DollarBidder类，否则机器人无法访问运行程序包。
• 所有方法应在500毫秒或更短的时间内返回。
• 机器人不需要是确定性的。
• 你的出价不需要是5美分的倍数。
• $1=100¢
• 调查结果将于2018年4月24日公布。

GitHub上的转轮

结果

在这里查看个别子弹。

MTargetedBot: $14.30
BuzzardBot: $9.83
BluffBot: $9.40
RiskRewardBot: $9.35
SecretBot: $8.50
LuckyDiceBot: $7.28
CounterBot: $6.05
MBot: $5.40
StackTraceObfuscaterBot: $5.20
EvilBot: $4.80
MarginalBot: $4.60
TargetValueBot: $4.59
InflationBot: $4.27
UpTo200: $4.20
InsiderTradingBot: $1.90
MimicBot: $1.50
BorkBorkBot: $1.22
DeterrentBot: $0.95
MarginalerBot: $0.00
RandBot: $-4.45
BreakEvenAsap: $-7.00
AnalystOptimizer: $-13.95
DeterredBot: $-1997.06
ScoreOverflowBot: $-21474844.15
MirrorBot: $-21475836.25

祝贺MTargetedBot获得14.30美元的利润！

Answer 1

MTargetedBot

public class MTargetedBot extends MBot {

    @Override
    protected int calcBid(int opponentsBid, boolean isPeeking, boolean isSubPeeking) {
        Class c = this.rivalClass;

        switch (c.getSimpleName()) {
            case "AnalystBot":
                if (isPeeking && !isSubPeeking) {
                    throw new RuntimeException();
                } else if (isPeeking) {
                    return 66666;
                }
                break;
            case "MirrorBot":
                if (isPeeking && !isSubPeeking) {
                    throw new RuntimeException();
                } else if (isPeeking) {
                    return 0;
                }
                break;
            case "GreedyBot":
            case "LuckyDiceBot":
            case "InflationBot":
            case "TargetValueBot":
                // not playing with ya
                return 0;
            case "MimicBot":
            case "BuzzardBot":
            case "MarginalBot":
            case "MarginalerBot":
            case "BluffBot":
            case "MBot":
                // go away, gimme easy money
                return isPeeking ? 66666 : 5;
            case "RandBot":
                // me or noone
                return 100;
            case "SecretBot":
                return 10;
            case "AnalystKiller":
            case "OnlyWinningMove":
            case "EvilBot":
            case "StackTraceObfuscaterBot":
                // easy
                return opponentsBid + 5;
        }

        return super.calcBid(opponentsBid, isPeeking, isSubPeeking);
    }
}

• 基于更新的MBot
• 使用类似的方法(如CounterBot )，但使用一些改进的方法来更有力地打击其中一些对手，也应该更容易读懂。
• 关于未知对手默认为MBot strat

Answer 2

MimicBot

import net.ramenchef.dollarauction.DollarBidder;

import java.util.Set;
import java.util.HashSet;

public class MimicBot extends AbstractAnalystCounterBot {

    private final Set> bidders = new HashSet<>();
    private DollarBidder reference = null;

    // A benchmark class. Not MarginalBot because of proposed rule changes.
    public static class BidFive extends DollarBidder {
        public int nextBid(int o) {
            return 5;
        }
    }


    public MimicBot() {
        bidders.add(OnlyWinningMove.class);
        bidders.add(GreedyBot.class);
        bidders.add(BidFive.class);
    }


    @Override
    public void newAuction(Class opponent) {
        DollarBidder enemy;
        reference = null;
        try {
            enemy = opponent.newInstance();
        } catch (Throwable t) {
            return;
        }

        if (!bidders.contains(opponent))
            bidders.add(opponent);

        Class leader = OnlyWinningMove.class;
        int best = 0;

        for (Class audition : bidders) {
            try {
                enemy.newAuction(MimicBot.class);
            } catch (Throwable t) {
                reference = new GreedyBot(); // Deterrence.
                break;
            }

            DollarBidder tryout;
            try {
                tryout = audition.newInstance();
                tryout.newAuction(opponent);
            } catch (Throwable t) {
                continue;
            }

            int tryoutScore = -100000;
            /* This code was copy-pasted from the *
             * runner, with significant changes. */
            int bid1 = 0, bid2 = 0;
            while (true) {
                int next;
                try {
                    next = enemy.nextBid(bid2);
                } catch (Throwable t) {
                    tryoutScore = 100;
                    break;
                }
                if (next < bid2 + 5) {
                    if (bid2 > 0) {
                        tryoutScore = 100 - bid1;
                    }
                    break;
                }
                if (next > 10000 && bid2 > 10000) {
                    tryoutScore = -10000;
                    break;
                }
                bid1 = next;

                try {
                    next = tryout.nextBid(bid1);
                } catch (Throwable t) {
                    tryoutScore = -bid2;
                    break;
                }
                if (next < bid1 + 5) {
                    tryoutScore = -bid2;
                    break;
                }
                if (next > 10000 && bid1 > 10000) {
                    tryoutScore = -10000;
                    break;
                }
                bid2 = next;
            }
            /* End of copy-pasted code. */

            if (tryoutScore > best) {
                best = tryoutScore;
                leader = audition;
            }
        }

        try {
            reference = leader.newInstance();
        } catch (Throwable t) {
            reference = new OnlyWinningMove();
        }
        reference.newAuction(opponent);
    }


    @Override
    public int nextBid(int opponentsBid) {
        try {
            return reference.nextBid(opponentsBid);
        } catch (Throwable t) {
            return 5;
        }
    }
}

天啊。我以为这是简单的写作，然后花了3个小时在上面。

本质上，MimicBot保存了可用的机器人的运行列表。当它进入一个新的拍卖，它运行在列表中寻找最有效的对抗当前的对手。然后，它使用该机器人作为“参考”在拍卖。

为了测试目的，最好是使用提交的随机子集或完整的集合。它以GreedyBot、MimicBot和另外一个仅出价5美分的机器人开始。

Answer 3

InsiderTradingBot

本着@StephenLeppik回答的精神，InsiderTradingBot了解他的所有对手并理解他们的策略。你的行动，斯蒂芬。

import net.ramenchef.dollarauction.DollarBidder;

public class InsiderTradingBot extends DollarBidder {
  private static boolean analystNutcracker = false;
  private int bid;

  @Override
  public void newAuction(Class opponent) {
    if (opponent.equals(DeterredBot.class) ||
        opponent.equals(OnlyWinningMove.class) ||
        opponent.equals(MirrorBot.class)) {
      // I can do this ^.^
      bid = 5;
    } else if (opponent.equals(AnalystKiller.class)) {
      // Outbid 'em >:D
      bid = 10;
    } else if (opponent.equals(BreakEvenAsap.class) ||
               opponent.equals(BorkBorkBot.class) ||
               opponent.equals(DeterrentBot.class)) {
      // Break even quicker!
      bid = 100;
    } else if (opponent.equals(InsiderTradingBot.class)) {
      // I'm probably a simulation inside MirrorBot
      bid = 0;
    } else if (opponent.equals(Analyst.class)) {
      // Let's fight the Analyst with the power of global variables
      bid = 100;
      analystNutcracker = true;
    } else {
      // Welp
      bid = 0;
    }
  }

  @Override
  public int nextBid(int opponentsBid) {
    if ((opponentsBid == 95) && analystNutcracker) {
      analystNutcracker = false;
      return 0;
    }
    return bid;
  }

};