feat: ISMCTS bot + full documentation
新增: - PdkFriendServer/Logic/IsmctsBot.cs: ISMCTS蒙特卡洛搜索机器人 - 800次模拟/决策,对手手牌从未知牌池随机采样 - 包庄决策:大牌(≥J/Q/K/A/2)数量 ≥6 则包庄 - winRate日志输出,决策透明 - docs/bot-architecture.md: 系统架构文档 - docs/bot-algorithms.md: A/B/C三条算法路径完整分析 - docs/ismcts-design.md: ISMCTS实现设计详案 切换: - PdkGameMain.InitBots: PdkBot → IsmctsBot(默认ISMCTS) 验证: - dotnet build 0 errors - 多人对局完整运行:包庄→出牌→结算→test over
This commit is contained in:
251
PdkFriendServer/Logic/IsmctsBot.cs
Normal file
251
PdkFriendServer/Logic/IsmctsBot.cs
Normal file
@ -0,0 +1,251 @@
|
||||
using System;
|
||||
using System.Collections.Generic;
|
||||
using System.Linq;
|
||||
using GameFix.PaoDeKuaiF;
|
||||
using GameFix.Poker;
|
||||
using GameMessage;
|
||||
using GameMessage.PaoDeKuaiF;
|
||||
|
||||
namespace PdkFriendServer.Logic
|
||||
{
|
||||
/// <summary>
|
||||
/// ISMCTS (Information Set Monte Carlo Tree Search) 机器人
|
||||
/// 对每个候选出牌方案模拟 N 局随机对局,选胜率最高的方案
|
||||
/// 只看到自己手牌+桌面信息——对手手牌从未知牌池随机采样
|
||||
/// </summary>
|
||||
public class IsmctsBot : IPdkBot
|
||||
{
|
||||
private readonly int _simulations;
|
||||
private readonly Random _rng;
|
||||
|
||||
public IsmctsBot(int simulations = 800)
|
||||
{
|
||||
_simulations = simulations;
|
||||
_rng = new Random();
|
||||
}
|
||||
|
||||
public bool DecideBaoZhuang(PdkBotView view)
|
||||
{
|
||||
// 包庄决策:用快速模拟评估——手牌质量高就包
|
||||
// 简化:手牌中 >10 的大牌(J/Q/K/A/2)多就包
|
||||
int bigCards = view.MyHand.Count(c => c.GameNum >= 11);
|
||||
return bigCards >= 6; // ≥6 张大牌 → 有一定控制力
|
||||
}
|
||||
|
||||
public PlayOutCardPdkF DecidePlay(PdkBotView view)
|
||||
{
|
||||
var hand = view.MyHand;
|
||||
if (hand.Length == 0)
|
||||
return Pass(view);
|
||||
|
||||
bool isFirstPlay = view.MaxPlayCard.GameNum <= 0;
|
||||
|
||||
// 枚举所有合法出牌
|
||||
var tips = PdkCardAlgorithm.GetTipCard(
|
||||
view.MaxPlayCard, hand, null, view.Rule.AAAIsZhaDan);
|
||||
|
||||
var candidates = new List<(PlayOutCardPdkF play, double winRate)>();
|
||||
if (!isFirstPlay)
|
||||
candidates.Add((Pass(view), 0));
|
||||
|
||||
if (tips != null)
|
||||
foreach (var tip in tips)
|
||||
candidates.Add((MakePlay(tip, view.MyPos), 0));
|
||||
|
||||
if (candidates.Count == 1)
|
||||
return candidates[0].play;
|
||||
|
||||
// 对每个候选做蒙特卡洛评估
|
||||
int unknownCount = 48 - hand.Length - CountPlayed(hand, view);
|
||||
int simsPerCandidate = Math.Max(50, _simulations / candidates.Count);
|
||||
|
||||
for (int ci = 0; ci < candidates.Count; ci++)
|
||||
{
|
||||
int wins = 0;
|
||||
var play = candidates[ci].play;
|
||||
|
||||
for (int s = 0; s < simsPerCandidate; s++)
|
||||
{
|
||||
if (SimulateOneGame(hand, play, view, unknownCount))
|
||||
wins++;
|
||||
}
|
||||
|
||||
candidates[ci] = (play, (double)wins / simsPerCandidate);
|
||||
}
|
||||
|
||||
candidates.Sort((a, b) => b.winRate.CompareTo(a.winRate));
|
||||
var best = candidates[0];
|
||||
|
||||
string desc = best.play.Type == CardType1.None ? "pass"
|
||||
: $"{best.play.Type}:{best.play.Ids?.Length ?? 0}c";
|
||||
Console.WriteLine($"[ISMCTS pos{view.MyPos}] {simsPerCandidate * candidates.Count}sims "
|
||||
+ $"winRate:{best.winRate:P0} => {desc}");
|
||||
|
||||
return best.play;
|
||||
}
|
||||
|
||||
// ---- 模拟引擎 ----
|
||||
|
||||
private bool SimulateOneGame(
|
||||
TCardInfoPdkF[] myHand, PlayOutCardPdkF myPlay,
|
||||
PdkBotView view, int unknownCount)
|
||||
{
|
||||
int myPos = view.MyPos;
|
||||
int nextPos = myPos % 3 + 1;
|
||||
int oppPos = (myPos + 1) % 3 + 1;
|
||||
|
||||
// 构建 3 人手牌:我已知,对手随机分配
|
||||
var hands = new List<TCardInfoPdkF>[3];
|
||||
hands[myPos - 1] = new List<TCardInfoPdkF>(myHand);
|
||||
hands[nextPos - 1] = new List<TCardInfoPdkF>();
|
||||
hands[oppPos - 1] = new List<TCardInfoPdkF>();
|
||||
|
||||
// 生成 48 张标准牌,排除我的手牌
|
||||
var allCards = GenerateDeck();
|
||||
var unknown = allCards.Where(c => !myHand.Any(h => h.ID == c.ID)).ToList();
|
||||
Shuffle(unknown);
|
||||
|
||||
int nextCnt = view.ShenYuCard[nextPos - 1];
|
||||
int oppCnt = view.ShenYuCard[oppPos - 1];
|
||||
hands[nextPos - 1].AddRange(unknown.Take(nextCnt));
|
||||
hands[oppPos - 1].AddRange(unknown.Skip(nextCnt).Take(oppCnt));
|
||||
|
||||
// 应用我的第一步出牌
|
||||
var maxPlay = new PlayOutCardPdkF
|
||||
{
|
||||
Pos = (byte)myPos,
|
||||
GameNum = myPlay.GameNum,
|
||||
Type = myPlay.Type,
|
||||
Ids = myPlay.Ids?.ToArray()
|
||||
};
|
||||
if (myPlay.Type != CardType1.None && myPlay.Ids != null)
|
||||
{
|
||||
foreach (var id in myPlay.Ids)
|
||||
hands[myPos - 1].RemoveAll(c => c.ID == id);
|
||||
}
|
||||
|
||||
// 模拟到结束
|
||||
int current = myPos % 3 + 1; // 下家先跟
|
||||
int maxPlayer = myPlay.Type == CardType1.None ? view.MaxPlayCard.Pos : myPos;
|
||||
if (myPlay.Type == CardType1.None)
|
||||
{
|
||||
maxPlayer = view.MaxPlayCard.Pos;
|
||||
maxPlay = CloneCard(view.MaxPlayCard);
|
||||
}
|
||||
|
||||
int moveCount = 0;
|
||||
int maxMoves = 500;
|
||||
|
||||
while (moveCount++ < maxMoves)
|
||||
{
|
||||
var curHand = hands[current - 1];
|
||||
if (curHand.Count == 0)
|
||||
return current == myPos; // 我赢了
|
||||
|
||||
bool isNewRound = maxPlayer == current;
|
||||
|
||||
// 获取当前玩家的合法出牌
|
||||
var curMax = isNewRound
|
||||
? new PlayOutCardPdkF() // GameNum=0 means new round
|
||||
: maxPlay;
|
||||
var curTips = PdkCardAlgorithm.GetTipCard(
|
||||
curMax, curHand.ToArray(), null, false);
|
||||
|
||||
if (isNewRound || (curTips != null && curTips.Count > 0))
|
||||
{
|
||||
// 选第一个方案(最小牌)
|
||||
var pick = isNewRound
|
||||
? curHand.OrderBy(c => c.GameNum).First()
|
||||
: curTips[0][0];
|
||||
var ids = isNewRound
|
||||
? new[] { pick.ID }
|
||||
: curTips[0].Select(c => c.ID).ToArray();
|
||||
|
||||
maxPlay = new PlayOutCardPdkF
|
||||
{
|
||||
Pos = (byte)current,
|
||||
GameNum = pick.GameNum,
|
||||
Type = curTips?[0].Count == 1 ? CardType1.DanZhang : CardType1.DuiZi,
|
||||
Ids = ids
|
||||
};
|
||||
maxPlayer = current;
|
||||
|
||||
foreach (var id in ids)
|
||||
curHand.RemoveAll(c => c.ID == id);
|
||||
}
|
||||
// else: pass, keep maxPlay unchanged
|
||||
|
||||
current = current % 3 + 1;
|
||||
}
|
||||
|
||||
// 超时——比较剩余牌数
|
||||
return hands[myPos - 1].Count <= hands[nextPos - 1].Count
|
||||
&& hands[myPos - 1].Count <= hands[oppPos - 1].Count;
|
||||
}
|
||||
|
||||
// ---- 工具 ----
|
||||
|
||||
private static int CountPlayed(TCardInfoPdkF[] myHand, PdkBotView view)
|
||||
{
|
||||
// 粗略估计已出牌:总牌 48 - 我的手牌 - 所有剩余牌
|
||||
int total = 48 - myHand.Length;
|
||||
foreach (var c in view.ShenYuCard) total -= c;
|
||||
return Math.Max(0, total);
|
||||
}
|
||||
|
||||
private static List<TCardInfoPdkF> GenerateDeck()
|
||||
{
|
||||
var deck = new List<TCardInfoPdkF>();
|
||||
int id = 1;
|
||||
// 花色 1=黑桃 2=红桃 3=梅花 4=方块, 大小 1-13 (A=14, 2=16)
|
||||
foreach (int flower in new[] { 1, 2, 3, 4 })
|
||||
for (int num = 1; num <= 13; num++)
|
||||
{
|
||||
int gameNum = num switch
|
||||
{
|
||||
1 => 14, // A
|
||||
2 => 16, // 2
|
||||
_ => num
|
||||
};
|
||||
deck.Add(new TCardInfoPdkF { ID = (byte)id++, Flower = (byte)flower, GameNum = (byte)gameNum, GameState = 1 });
|
||||
}
|
||||
return deck;
|
||||
}
|
||||
|
||||
private void Shuffle<T>(List<T> list)
|
||||
{
|
||||
for (int i = list.Count - 1; i > 0; i--)
|
||||
{
|
||||
int j = _rng.Next(i + 1);
|
||||
(list[i], list[j]) = (list[j], list[i]);
|
||||
}
|
||||
}
|
||||
|
||||
private static PlayOutCardPdkF CloneCard(PlayOutCardPdkF src)
|
||||
{
|
||||
return new PlayOutCardPdkF
|
||||
{
|
||||
Pos = src.Pos,
|
||||
GameNum = src.GameNum,
|
||||
Type = src.Type,
|
||||
Ids = src.Ids?.ToArray()
|
||||
};
|
||||
}
|
||||
|
||||
private static PlayOutCardPdkF MakePlay(List<TCardInfoPdkF> cards, int pos)
|
||||
{
|
||||
return new PlayOutCardPdkF
|
||||
{
|
||||
Pos = (byte)pos,
|
||||
GameNum = cards[0].GameNum,
|
||||
Ids = cards.Select(x => x.ID).ToArray(),
|
||||
Type = CardType1.DanZhang
|
||||
};
|
||||
}
|
||||
|
||||
private static PlayOutCardPdkF Pass(PdkBotView view)
|
||||
{
|
||||
return new PlayOutCardPdkF { Pos = (byte)view.MyPos, Type = CardType1.None };
|
||||
}
|
||||
}
|
||||
}
|
||||
@ -61,7 +61,7 @@ namespace PdkFriendServer.Logic
|
||||
{
|
||||
_bots = new IPdkBot[3];
|
||||
for (int i = 0; i < 3; i++)
|
||||
_bots[i] = new PdkBot();
|
||||
_bots[i] = new IsmctsBot();
|
||||
}
|
||||
|
||||
/// <summary>构建当前玩家的可见视图——只看自己手牌+公开信息</summary>
|
||||
|
||||
Reference in New Issue
Block a user