[MCTS] 不围棋NoGoAI

计概课的大作业

学了一下MCTS，随便调了下参数（

upd：被锤爆了 需要修改一下随机后的估值（咕咕咕

---

MCTS参考资料

https://blog.csdn.net/baidu_40614951/article/details/105481498

https://www.cnblogs.com/yifdu25/p/8303462.html

界面相关

https://blog.csdn.net/luoshengkim/article/details/50412354

平台

https://www.botzone.org.cn/game/NoGo

---

扔一下代码

#include <cstdio>
#include <cstring>
#include <algorithm>
#include <cmath>
#include <vector>
#include <string>
#include <iostream>
#include "jsoncpp/json.h"
using namespace std;
const int N=9;
const int M=10000;
const int dx[4]={-1,0,1,0},dy[4]={0,-1,0,1};
int currBotColor;

class Grid {
public:
    int grid[N][N],avasz,curCol,lose;//,turn=1; //turn 记为我方已落子数+1
    bool use[N*N];
    vector<int> ava;
    bool vis[N][N];
    bool ingrid(int x,int y) {return x>=0 && y>=0 && x<9 && y<9;}
    bool dfsqi(int x,int y) { //判断是否有气
        int tx,ty; vis[x][y]=1; bool flag=0;
        for (int k=0; k<4; k++) if (ingrid(tx=x+dx[k],ty=y+dy[k])) {
            if (!grid[tx][ty]) flag=1;
            else if (grid[x][y]==grid[tx][ty] && !vis[tx][ty] && dfsqi(tx,ty)) flag=1;
        }
        return flag;
    }
    bool judgeAvailable(int x,int y,int col) {
        if (grid[x][y]) return 0;
        grid[x][y]=col; memset(vis,0,sizeof vis);
        if (!dfsqi(x,y)) {grid[x][y]=0; return 0;}
        int tx,ty;
        for (int k=0; k<4; k++) if (ingrid(tx=x+dx[k],ty=y+dy[k])&&!vis[tx][ty]) 
            if (col==-grid[tx][ty]&&!dfsqi(tx,ty)) {grid[x][y]=0;return 0;}
        grid[x][y]=0; return 1;
    }
    void checkavailable() {
        if (ava.size()) ava.clear();
        for (int i=0; i<9; i++) for (int j=0; j<9; j++) if (judgeAvailable(i,j,curCol))
            ava.push_back(i*9+j);
        avasz=ava.size(); if (!avasz) lose=1; else lose=0;
        memset(use,0,sizeof use);
    }
    int getaction() {
        if (!avasz) return -1;
        return ava[rand()%avasz];
    }
} G;

// const double Con = 0.70710678; //调参
const double Con = 0.8;
const int SearchDepth = 60;
const int TimeLimit = 15000;
int cnt,rt,act[M],A[M],B[M],fa[M];// A/B表胜率
Grid g[M];
vector<int> son[M];

int BestChild(int v,bool flag,double c) {//c是参数
    int id; double mx=-1,val;
    if (g[v].curCol != currBotColor) {
        if (flag) for (int i=son[v].size()-1; ~i; i--) {
            if (!B[son[v][i]]) val=1e5;
            else val=1.0*(B[son[v][i]]-A[son[v][i]])/B[son[v][i]]+c*sqrt(2*log(B[v])/B[son[v][i]]);//需要保证B非零！！
            if (val-mx>1e-8) mx=val,id=i;
        } else for (int i=son[v].size()-1; ~i; i--) {
            if (!B[son[v][i]]) val=1e5;
            else val=1.0*(B[son[v][i]]-A[son[v][i]])/B[son[v][i]];//需要保证B非零！！
            if (val-mx>1e-8) mx=val,id=i;
        }
    } else {
        if (flag) for (int i=son[v].size()-1; ~i; i--) {
            if (!B[son[v][i]]) val=1e5;
            else val=1.0*A[son[v][i]]/B[son[v][i]]+c*sqrt(2*log(B[v])/B[son[v][i]]);//需要保证B非零！！
            if (val-mx>1e-8) mx=val,id=i;
        } else for (int i=son[v].size()-1; ~i; i--) {
            if (!B[son[v][i]]) val=1e5;
            else val=1.0*A[son[v][i]]/B[son[v][i]];//需要保证B非零！！
            if (val-mx>1e-8) mx=val,id=i;
        }
    }
    return son[v][id];
}
int expand(int v) {// Expansion
    for (int i=son[v].size()-1; ~i; i--) g[v].use[act[son[v][i]]]=1;
    vector<int> tmp;
    for (int i=g[v].avasz-1; ~i; i--) if (!g[v].use[g[v].ava[i]]) tmp.push_back(g[v].ava[i]);
    int ac = tmp[rand()%tmp.size()];
    g[++cnt]=g[v],g[cnt].grid[ac/9][ac%9]=g[v].curCol,g[cnt].curCol=-g[v].curCol;
    fa[cnt]=v; son[v].push_back(cnt); act[cnt]=ac; g[cnt].checkavailable();
    return cnt;
}
int TreePolicy(int v) {
    int dep=SearchDepth;
    while (g[v].avasz && dep--) //调参（控制层数）
        if (son[v].size()<g[v].avasz) return expand(v);
        else v=BestChild(v,1,Con);
    return v;
}
int DefaultPolicy(int v) {
    Grid gt=g[v];
    vector<int> a,b; int pa=0,pb=0; bool cur=1;
    for (int i=0; i<81; i++) if (!gt.grid[i/9][i%9]) a.push_back(i),b.push_back(i);
    for (int i=0; i<a.size(); i++) swap(a[i],a[rand()%a.size()]);
    for (int i=0; i<b.size(); i++) swap(b[i],b[rand()%b.size()]);
    while (pa<a.size()&&pb<b.size()) {
        if (cur) {
            while (pa<a.size() && !gt.judgeAvailable(a[pa]/9,a[pa]%9,gt.curCol)) pa++;
            if (pa==a.size()) return currBotColor == gt.curCol ? 0:1;
            gt.grid[a[pa]/9][a[pa]%9]=gt.curCol,pa++,cur=0;
        }
        else {
            while (pb<b.size() && !gt.judgeAvailable(b[pb]/9,b[pb]%9,-gt.curCol)) pb++;
            if (pb==b.size()) return currBotColor == -gt.curCol ? 0:1;
            gt.grid[b[pb]/9][b[pb]%9]=-gt.curCol,pb++,cur=1;
        }
    }
    if (pa==a.size()) return currBotColor == gt.curCol ? 0:1;
    else return currBotColor == -gt.curCol ? 0:1;
}
inline void BackUp(int v,int dt) {for (; v; v=fa[v]) A[v]+=dt,B[v]++;} //判断胜负以currBotColor计
int MCTS(int runtime) {
    cnt=rt=1; g[1]=G;
    while (runtime--) {
        int v=TreePolicy(rt); //Selection
        int dt=DefaultPolicy(v); //Simulation
        BackUp(v,dt); //Backpropagation
    }
    return act[BestChild(rt,0,0)];
}

int main() {
    srand((unsigned)time(0));
    string str;
    getline(cin,str);
    Json::Reader reader;
    Json::Value input;
    reader.parse(str, input);

    int x,y;
    int turnID = input["responses"].size();
    G.curCol = currBotColor = input["requests"][0]["x"].asInt()<0 ? 1:-1;
    for (int i = 0; i < turnID; i++) {
        x=input["requests"][i]["x"].asInt(), y=input["requests"][i]["y"].asInt();
        if (x!=-1) G.grid[x][y]=-currBotColor;
        x=input["responses"][i]["x"].asInt(), y=input["responses"][i]["y"].asInt();
        if (x!=-1) G.grid[x][y]=currBotColor;//,turn++;
    }
    x=input["requests"][turnID]["x"].asInt(), y=input["requests"][turnID]["y"].asInt();
    if (x!=-1) G.grid[x][y]=-currBotColor;
    
    G.checkavailable();
    int decision=MCTS(TimeLimit);//调参
    
    Json::Value ret;
    Json::Value action;
    action["x"]=decision/9; action["y"]=decision%9;
    ret["response"] = action;
    Json::FastWriter writer;
    cout << writer.write(ret) << endl;
    return 0;
}