1. 程式人生 > >DLX演算法及應用(一)DLX模板+解數獨

DLX演算法及應用(一)DLX模板+解數獨

DLX演算法

原理:網上太多了,我就不寫了。。

用途:解決精確覆蓋問題

下面的程式碼是嚴格按照演算法寫的,其實對於這種沒有資料域的連結串列,是可以用陣列進行模擬的(見DLX演算法及應用(二)Matlab解數獨)。

程式碼中全部都用的是vector,更通用一些~

後半部分給出了一個解數獨的例項,如何將數獨轉換為精確覆蓋問題的文章也是網上一搜一大把。。。。我就不寫了。。

2014-2-15 修改:還是寫下如何轉換吧

精確覆蓋問題的矩陣表示:

給出這樣一個矩陣,

1	0	0	1	0	0	1
1	0	0	1	0	0	0
0	0	0	1	1	0	1
0	0	1	0	1	1	0
0	1	1	0	0	1	1
0	1	0	0	0	0	1
找出該矩陣一組行的集合,使得該組集合中每列有且只有一個1

例如第2,4,6行

那麼我們現在將數獨問題轉換成一個324列的精確覆蓋問題:

數獨有4個限制條件,將其分別轉換為81列

每個格子都要填入數字
---1到81列,表示數獨中9*9=81個格子是否填入了數字。如果是,則選取的01行在該01列上為1

每一行都要有1~9填入
---81+1到81*2列,每9列就代表數獨中的一行,如果該行有某個數字,則其對應的列上為1

每一列都要有1~9填入
---81*2+1到81*3列,每9列就代表數獨中的一列

每一宮都要有1~9填入
---81*3+1到81*4列,每9列就代表數獨中的一宮
那讓我們從第一個格子開始,到第81個格子,逐步構造01矩陣

對於已給出數字的格子,例如第 3 行第 5 列為7

那麼就插入一行,其中,第23,106,205,259列為1,其他為0,分別代表:

23 =        (3-1)*9 + 5    第 3 行第 5 列填入了數字
106= 81   + (3-1)*9 + 7    第 3 行填入了7
205= 81*2 + (5-1)*9 + 7    第 5 列填入了7
259= 81*3 + (2-1)*9 + 7    第 2 宮填入了7

如果沒有給出數字,那這個格子就有9種可能性,插入9行,其中如下列為1:
第一行:23,100,199,253
第二行:23,101,200,254
........
第九行:23,108,207,261
這樣,構造的01矩陣,每行都有4個1。

在最多81*9行的01矩陣中,尋找一組精確覆蓋(81行),就可以求解一個數獨

程式碼如下

#include <time.h>
#include <iostream>
#include <limits.h>
#include <vector>
#include <fstream>
using namespace std;

struct Node{
    Node *up, *down, *left, *right, *colRoot, *rowRoot;//上下左右四個指標以及指向行列物件的指標
    int Num;//行物件特有,記錄行數
    int Size;//列物件特有,記錄該列元素數
    Node(int i = -1 ): Num(i),Size(0) {};//建構函式
};

class DLX{
public:
    DLX(vector<vector<int> > &matrix, int m, int n);
    ~DLX() { delete Head;};//析構有點難寫
    void init();
    void link(vector<vector<int> > &matrix);
    void cover(Node *cRoot);
    void recover(Node *cRoot);
    bool Search(int k = 0);
    vector<int> getResult() const { return result;}
    int getUpdates() const { return _updates;}
private:
    Node *Head;
    vector<int> result;//結果存放在這裡
    int _row, _col, _updates;//記錄行列數,更新次數
};

DLX::DLX(vector<vector<int> > &matrix, int m, int n)
    :_row(m),_col(n),_updates(0)
{
    Head = new Node;
    Head->up = Head;
    Head->down = Head;
    Head->right = Head;
    Head->left = Head;
    init();
    link(matrix);
}

void DLX::init()
{
    Node *newNode;
    for (int ix = 0; ix < _col; ++ix)//表頭位置向後插入,構造列物件
    {
        newNode = new Node;
        newNode->up = newNode;
        newNode->down = newNode;
        newNode->right = Head->right;
        newNode->left = Head;
        newNode->right->left = newNode;
        Head->right = newNode;
    }
    for (int ix = 0; ix < _row; ++ix)//表頭位置向下插入,構造行物件
    {
        newNode = new Node(_row-ix);//注意序號是_row-ix
        newNode->down = Head->down;
        newNode->up = Head;
        newNode->down->up = newNode;
        Head->down = newNode;
    }
}

void DLX::link(vector<vector<int> > &matrix)
{
    Node *current_row, *current_col, *newNode, *current;//當前行物件,當前列物件,新節點,當前節點
    current_row = Head;
    for (int row = 0; row < _row; ++row)
    {
        current_row = current_row->down;
        current_col = Head;
        for (int col = 0; col < _col; ++col)
        {
            current_col = current_col->right;

            if (matrix[row][col] == 0)//矩陣上為0的位置不設定節點
                continue;

            newNode = new Node;

            newNode->colRoot = current_col;
            newNode->rowRoot = current_row;//設定當前節點對應的行列物件

            newNode->down = current_col;
            newNode->up = current_col->up;
            newNode->up->down = newNode;
            current_col->up = newNode;//連結當前節點到列雙向鏈尾端

            if (current_row->Size == 0)//行雙向鏈不應該把行物件包含進來
            {
                current_row->right = newNode;
                newNode->left = newNode;
                newNode->right = newNode;
                current_row->Size++;
            }
            current = current_row->right;//設定當前節點(即行物件右的節點)
            newNode->left = current->left;
            newNode->right = current;
            newNode->left->right = newNode;
            current->left = newNode;//連結當前節點到行雙向鏈尾端

            current_col->Size++;
        }
    }
}

void DLX::cover(Node *cRoot)//覆蓋列
{
    ++_updates;
    cRoot->left->right = cRoot->right;
    cRoot->right->left = cRoot->left;//刪除該列物件
    Node *i, *j;
    i = cRoot->down;
    while (i != cRoot)
    {
        j = i->right;
        while (j != i)
        {
            j->down->up = j->up;
            j->up->down = j->down;
            j->colRoot->Size--;
            j = j->right;
        }
        i = i->down;
    }
}

void DLX::recover(Node *cRoot)//整個演算法的精髓!!
{
    Node *i, *j;
    i = cRoot->up;
    while (i != cRoot)
    {
        j = i->left;
        while (j != i)
        {
            j->colRoot->Size++;
            j->down->up = j;
            j->up->down = j;
            j = j->left;
        }
        i = i->up;
    }
    cRoot->right->left = cRoot;
    cRoot->left->right = cRoot;
}

bool DLX::Search(int k)
{
    if (Head->right == Head)//表空,則成功找到一組行的集合
        return true;

    Node *cRoot, *c;
    int minSize = INT_MAX;
    for(c = Head->right; c != Head; c = c->right)//根據啟發條件選擇列物件
    {
        if (c->Size < minSize)
        {
            minSize = c->Size;
            cRoot = c;
            if (minSize == 1)
                break;
            if (minSize == 0)//有一列為空,失敗
                return false;
        }
    }
    cover(cRoot);

    Node *current_row,*current;
    for (current_row = cRoot->down; current_row != cRoot; current_row = current_row->down)
    {
        result.push_back(current_row->rowRoot->Num);//將該行加入result中
        for (current = current_row->right; current != current_row; current = current->right)
        {
            cover(current->colRoot);
        }
        if (Search(k+1))
            return true;
        for (current = current_row->left; current != current_row; current = current->left)
            recover(current->colRoot);
        result.pop_back();//發現該行不符合要求,還原result
    }
    recover(cRoot);
    return false;
}

vector<vector<int> > sudoku2matrix(string &problem)//將數獨轉換為01矩陣
{
    vector<vector<int> > matrix;
    for (int ix = 0; ix < 81; ++ix)
    {
        int val = problem[ix] - '0';
        vector<int> current_row(324,0);
        if (val != 0)
        {
            current_row[ix] = 1;
            current_row[81 + ix/9*9 + val -1] = 1;
            current_row[162 + ix%9*9 +val -1] = 1;
            current_row[243 + (ix/9/3*3+ix%9/3)*9 +val -1] = 1;
            matrix.push_back(current_row);
            continue;
        }
        for (int jx = 0; jx < 9; ++jx)
        {
            vector<int> current_row2(324,0);
            current_row2[ix] = 1;
            current_row2[81 + ix/9*9 + jx] = 1;
            current_row2[162 + ix%9*9 +jx] = 1;
            current_row2[243 + (ix/9/3*3+ix%9/3)*9 +jx] = 1;
            matrix.push_back(current_row2);
        }
    }
    return matrix;
}

vector<int> matrix2sudoku(vector<vector<int> > &matrix, vector<int> result)//將01矩陣翻譯為數獨
{
    vector<int> solution(81);
    for (int ix = 0; ix < 81; ++ix)
    {
        vector<int> current = matrix[result[ix]-1];
        int pos = 0, val = 0;
        for (int jx = 0; jx < 81; ++jx)
        {
            if (current[jx] == 1)
                break;
            ++pos;
        }
        for (int kx = 81; kx < 162; ++kx)
        {
            if (current[kx] == 1)
                break;
            ++val;
        }
        solution[pos] = val%9 + 1;
    }
    return solution;
}

void solve_sudoku(string &problem, ostream &os = cout)
{
    clock_t start_1 = clock();
    vector<vector<int> > matrix = sudoku2matrix(problem);
    clock_t end_1 = clock();
    float time_1=(float)(end_1-start_1)/CLOCKS_PER_SEC;

    clock_t start_2 = clock();
    DLX sudoku(matrix,matrix.size(),324);
    clock_t end_2 = clock();
    float time_2=(float)(end_2-start_2)/CLOCKS_PER_SEC;

    clock_t start_3 = clock();
    if (!sudoku.Search())
    {
        os << "該數獨無解!\n\n";
        return;
    }
    clock_t end_3 = clock();
    float time_3=(float)(end_3-start_3)/CLOCKS_PER_SEC;

    clock_t start_4 = clock();
    vector<int> solution = matrix2sudoku(matrix, sudoku.getResult());
    clock_t end_4 = clock();
    float time_4=(float)(end_4-start_4)/CLOCKS_PER_SEC;

    for (int ix = 0; ix < 81; ++ix)
        os << solution[ix] << ((ix+1)%9 ? '\0' : '\n');

    os << "構造01矩陣用時: " << time_1 << "s\n"
         << "構造連結串列用時: " << time_2 << "s\n"
         << "Dancing用時: " << time_3 << "s\n"
         << "Dancing更新次數: " << sudoku.getUpdates() << "次\n"
         << "翻譯結果用時: " << time_4 << "s\n" << endl;
}

int main()
{
    string problem;
    ofstream outfile("solution.txt");
    ifstream infile("problem.txt");
    while (infile >> problem)
    {
        outfile << problem << endl;
        if (problem.size() != 81)
        {
            outfile << "數獨不合法\n\n";
            continue;
        }
        solve_sudoku(problem, outfile);
    }
}


示例:

problem.txt檔案內容:
027380010010006735000000029305692080000000000060174503640000000951800070080065340
000000520080400000030009000501000600200700000000300000600010000000000704000000030
800000000003600000070090200050007000000045700000100030001000068008500010090000400

執行後生成的solution.txt內容
027380010010006735000000029305692080000000000060174503640000000951800070080065340
5 2 7 3 8 9 4 1 6
8 1 9 4 2 6 7 3 5
4 3 6 7 5 1 8 2 9
3 7 5 6 9 2 1 8 4
1 9 4 5 3 8 2 6 7
2 6 8 1 7 4 5 9 3
6 4 3 2 1 7 9 5 8
9 5 1 8 4 3 6 7 2
7 8 2 9 6 5 3 4 1
構造01矩陣用時: 0.002s
構造連結串列用時: 0.002s
Dancing用時: 0s
Dancing更新次數: 324次
翻譯結果用時: 0s

000000520080400000030009000501000600200700000000300000600010000000000704000000030
4 1 6 8 3 7 5 2 9
9 8 2 4 6 5 3 7 1
7 3 5 1 2 9 4 6 8
5 7 1 2 9 8 6 4 3
2 9 3 7 4 6 1 8 5
8 6 4 3 5 1 2 9 7
6 4 7 9 1 3 8 5 2
3 5 9 6 8 2 7 1 4
1 2 8 5 7 4 9 3 6
構造01矩陣用時: 0.002s
構造連結串列用時: 0.002s
Dancing用時: 0s
Dancing更新次數: 419次
翻譯結果用時: 0s

800000000003600000070090200050007000000045700000100030001000068008500010090000400
8 1 2 7 5 3 6 4 9
9 4 3 6 8 2 1 7 5
6 7 5 4 9 1 2 8 3
1 5 4 2 3 7 8 9 6
3 6 9 8 4 5 7 2 1
2 8 7 1 6 9 5 3 4
5 2 1 9 7 4 3 6 8
4 3 8 5 2 6 9 1 7
7 9 6 3 1 8 4 5 2
構造01矩陣用時: 0.002s
構造連結串列用時: 0.002s
Dancing用時: 0.001s
Dancing更新次數: 8321次
翻譯結果用時: 0s


最後一個數獨就是傳說中芬蘭數學家設計的最難數獨,可以從更新次數看出,確實挺難。。

整個演算法時間都浪費在了將數獨轉換成01矩陣和構造連結串列上,真正的DLX用時還是很短的!

(DLX演算法中一次覆蓋列的操作稱為一次更新)