N皇后是将N个象棋皇后放在N×N棋盘上的问题,这样就没有两个皇后互相攻击。
例如,以下是8 Queen 问题的解决方案。
Input: N = 4
Output:
0 1 0 0
0 0 0 1
1 0 0 0
0 0 1 0
Explanation:
The Position of queens are:
1 – {1, 2}
2 – {2, 4}
3 – {3, 1}
4 – {4, 3}
As we can see that we have placed all 4 queens
in a way that no two queens are attacking each other.
So, the output is correct
Input: N = 8
Output:
0 0 0 0 0 0 1 0
0 1 0 0 0 0 0 0
0 0 0 0 0 1 0 0
0 0 1 0 0 0 0 0
1 0 0 0 0 0 0 0
0 0 0 1 0 0 0 0
0 0 0 0 0 0 0 1
0 0 0 0 1 0 0 0
方法:想法是使用爬山算法。
- 虽然有像 Backtracking 这样的算法来解决 N Queen 问题,但让我们采用 AI 方法来解决这个问题。
- 很明显,AI 并不能一直保证全局正确的解决方案,但它有相当不错的成功率,约为 97%,这还不错。
- 将给出问题中使用的所有术语的概念的描述,如下所示:-
- 状态的概念——这里的状态是 NXN 板上 N 个皇后的任何配置。此外,为了减少搜索空间,让我们添加一个额外的约束,即在特定列中只能有一个皇后。程序中的一个状态是使用长度为 N 的数组实现的,这样如果state[i]=j则在列索引i和行索引j处有一个皇后。
- 邻居的概念– 一个州的邻居是其他州,其棋盘配置与当前状态的棋盘配置不同,仅就一个女王的位置而言。这个与其邻居不同状态的女王可能会被转移到同一列中的任何地方。
- 优化函数或目标函数——我们知道局部搜索是一种优化算法,它搜索局部空间来优化一个以状态为输入并给出一些值作为输出的函数。在这种情况下,状态的目标函数的值是相互攻击的皇后对的数量。我们的目标是找到具有最小目标值的状态。该函数的最大值为 NC2,最小值为 0。
算法:
- 从随机状态(即电路板的随机配置)开始。
- 扫描当前状态的所有可能邻居并跳转到具有最高目标值的邻居(如果有)。如果不存在目标严格高于当前状态的邻居,但存在一个相等的邻居,则跳转到任何随机邻居(逃避肩部和/或局部最优)。
- 重复第 2 步,直到找到一个目标严格高于其所有邻居目标的状态,然后转到第 4 步。
- 这样在局部搜索后找到的状态要么是局部最优,要么是全局最优。没有办法逃避局部最优,但是每次遇到局部最优时添加一个随机邻居或随机重启会增加实现全局最优(我们问题的解决方案)的机会。
- 输出状态并返回。
- 很容易看出,在我们的情况下,全局最优值为 0,因为它是可以相互攻击的皇后对的最小数量。此外,随机重启有更高的机会实现全局最优,但我们仍然使用随机邻居,因为我们的 N 个皇后问题没有大量的局部最优并且随机邻居比随机重启更快。
- 结论:
- Random Neighbor逃脱了肩膀,但只有很小的机会逃脱了局部最优。
- Random Restart既可以逃避肩膀,也很有可能逃避局部最优。
下面是爬山算法的实现:
CPP
// C++ implementation of the
// above approach
#include
#include
#define N 8
using namespace std;
// A utility function that configures
// the 2D array "board" and
// array "state" randomly to provide
// a starting point for the algorithm.
void configureRandomly(int board[][N],
int* state)
{
// Seed for the random function
srand(time(0));
// Iterating through the
// column indices
for (int i = 0; i < N; i++) {
// Getting a random row index
state[i] = rand() % N;
// Placing a queen on the
// obtained place in
// chessboard.
board[state[i]][i] = 1;
}
}
// A utility function that prints
// the 2D array "board".
void printBoard(int board[][N])
{
for (int i = 0; i < N; i++) {
cout << " ";
for (int j = 0; j < N; j++) {
cout << board[i][j] << " ";
}
cout << "\n";
}
}
// A utility function that prints
// the array "state".
void printState(int* state)
{
for (int i = 0; i < N; i++) {
cout << " " << state[i] << " ";
}
cout << endl;
}
// A utility function that compares
// two arrays, state1 and state2 and
// returns true if equal
// and false otherwise.
bool compareStates(int* state1,
int* state2)
{
for (int i = 0; i < N; i++) {
if (state1[i] != state2[i]) {
return false;
}
}
return true;
}
// A utility function that fills
// the 2D array "board" with
// values "value"
void fill(int board[][N], int value)
{
for (int i = 0; i < N; i++) {
for (int j = 0; j < N; j++) {
board[i][j] = value;
}
}
}
// This function calculates the
// objective value of the
// state(queens attacking each other)
// using the board by the
// following logic.
int calculateObjective(int board[][N],
int* state)
{
// For each queen in a column, we check
// for other queens falling in the line
// of our current queen and if found,
// any, then we increment the variable
// attacking count.
// Number of queens attacking each other,
// initially zero.
int attacking = 0;
// Variables to index a particular
// row and column on board.
int row, col;
for (int i = 0; i < N; i++) {
// At each column 'i', the queen is
// placed at row 'state[i]', by the
// definition of our state.
// To the left of same row
// (row remains constant
// and col decreases)
row = state[i], col = i - 1;
while (col >= 0
&& board[row][col] != 1) {
col--;
}
if (col >= 0
&& board[row][col] == 1) {
attacking++;
}
// To the right of same row
// (row remains constant
// and col increases)
row = state[i], col = i + 1;
while (col < N
&& board[row][col] != 1) {
col++;
}
if (col < N
&& board[row][col] == 1) {
attacking++;
}
// Diagonally to the left up
// (row and col simoultaneously
// decrease)
row = state[i] - 1, col = i - 1;
while (col >= 0 && row >= 0
&& board[row][col] != 1) {
col--;
row--;
}
if (col >= 0 && row >= 0
&& board[row][col] == 1) {
attacking++;
}
// Diagonally to the right down
// (row and col simoultaneously
// increase)
row = state[i] + 1, col = i + 1;
while (col < N && row < N
&& board[row][col] != 1) {
col++;
row++;
}
if (col < N && row < N
&& board[row][col] == 1) {
attacking++;
}
// Diagonally to the left down
// (col decreases and row
// increases)
row = state[i] + 1, col = i - 1;
while (col >= 0 && row < N
&& board[row][col] != 1) {
col--;
row++;
}
if (col >= 0 && row < N
&& board[row][col] == 1) {
attacking++;
}
// Diagonally to the right up
// (col increases and row
// decreases)
row = state[i] - 1, col = i + 1;
while (col < N && row >= 0
&& board[row][col] != 1) {
col++;
row--;
}
if (col < N && row >= 0
&& board[row][col] == 1) {
attacking++;
}
}
// Return pairs.
return (int)(attacking / 2);
}
// A utility function that
// generates a board configuration
// given the state.
void generateBoard(int board[][N],
int* state)
{
fill(board, 0);
for (int i = 0; i < N; i++) {
board[state[i]][i] = 1;
}
}
// A utility function that copies
// contents of state2 to state1.
void copyState(int* state1, int* state2)
{
for (int i = 0; i < N; i++) {
state1[i] = state2[i];
}
}
// This function gets the neighbour
// of the current state having
// the least objective value
// amongst all neighbours as
// well as the current state.
void getNeighbour(int board[][N],
int* state)
{
// Declaring and initializing the
// optimal board and state with
// the current board and the state
// as the starting point.
int opBoard[N][N];
int opState[N];
copyState(opState,
state);
generateBoard(opBoard,
opState);
// Initializing the optimal
// objective value
int opObjective
= calculateObjective(opBoard,
opState);
// Declaring and initializing
// the temporary board and
// state for the purpose
// of computation.
int NeighbourBoard[N][N];
int NeighbourState[N];
copyState(NeighbourState,
state);
generateBoard(NeighbourBoard,
NeighbourState);
// Iterating through all
// possible neighbours
// of the board.
for (int i = 0; i < N; i++) {
for (int j = 0; j < N; j++) {
// Condition for skipping the
// current state
if (j != state[i]) {
// Initializing temporary
// neighbour with the
// current neighbour.
NeighbourState[i] = j;
NeighbourBoard[NeighbourState[i]][i]
= 1;
NeighbourBoard[state[i]][i]
= 0;
// Calculating the objective
// value of the neighbour.
int temp
= calculateObjective(
NeighbourBoard,
NeighbourState);
// Comparing temporary and optimal
// neighbour objectives and if
// temporary is less than optimal
// then updating accordingly.
if (temp <= opObjective) {
opObjective = temp;
copyState(opState,
NeighbourState);
generateBoard(opBoard,
opState);
}
// Going back to the original
// configuration for the next
// iteration.
NeighbourBoard[NeighbourState[i]][i]
= 0;
NeighbourState[i] = state[i];
NeighbourBoard[state[i]][i] = 1;
}
}
}
// Copying the optimal board and
// state thus found to the current
// board and, state since c++ doesn't
// allow returning multiple values.
copyState(state, opState);
fill(board, 0);
generateBoard(board, state);
}
void hillClimbing(int board[][N],
int* state)
{
// Declaring and initializing the
// neighbour board and state with
// the current board and the state
// as the starting point.
int neighbourBoard[N][N] = {};
int neighbourState[N];
copyState(neighbourState, state);
generateBoard(neighbourBoard,
neighbourState);
do {
// Copying the neighbour board and
// state to the current board and
// state, since a neighbour
// becomes current after the jump.
copyState(state, neighbourState);
generateBoard(board, state);
// Getting the optimal neighbour
getNeighbour(neighbourBoard,
neighbourState);
if (compareStates(state,
neighbourState)) {
// If neighbour and current are
// equal then no optimal neighbour
// exists and therefore output the
// result and break the loop.
printBoard(board);
break;
}
else if (calculateObjective(board,
state)
== calculateObjective(
neighbourBoard,
neighbourState)) {
// If neighbour and current are
// not equal but their objectives
// are equal then we are either
// approaching a shoulder or a
// local optimum, in any case,
// jump to a random neighbour
// to escape it.
// Random neighbour
neighbourState[rand() % N]
= rand() % N;
generateBoard(neighbourBoard,
neighbourState);
}
} while (true);
}
// Driver code
int main()
{
int state[N] = {};
int board[N][N] = {};
// Getting a starting point by
// randomly configuring the board
configureRandomly(board, state);
// Do hill climbing on the
// board obtained
hillClimbing(board, state);
return 0;
}
输出:
0 0 1 0 0 0 0 0
0 0 0 0 0 1 0 0
0 0 0 0 0 0 0 1
1 0 0 0 0 0 0 0
0 0 0 1 0 0 0 0
0 0 0 0 0 0 1 0
0 0 0 0 1 0 0 0
0 1 0 0 0 0 0 0
复杂性分析
- 该算法的时间复杂度可以分为三个部分:
- 计算目标–目标的计算涉及遍历船上的所有皇后并检查编号。攻击皇后,这是由我们的calculateObjective函数在O(N 2 )时间内完成的。
- 邻居选择和邻居数量——我们问题中邻居的描述给出了当前状态的总共N(N-1) 个邻居。选择过程是最合适的,因此需要遍历所有邻居,这也是O(N 2 ) 。
- 搜索空间——我们问题的搜索空间总共包含N N 状态,对应于船上N Queens 的所有可能配置。请注意,这是在考虑到每列一个皇后的附加约束之后。
- 因此,我们算法的最坏情况时间复杂度是O(N N ) 。但是,这种最坏的情况在实践中很少发生,因此我们可以放心地认为它与处理 N Queen 问题的任何其他算法一样好。因此,有效时间复杂度只包括计算所有邻居的目标直到某个深度(搜索进行的跳跃次数),这不依赖于N。因此,如果搜索深度为d,则时间复杂度为O(N 2 * N 2 * d) ,即O(d*N 4 )。