Random tech excursions

N-Queens Problem and the JuMP Package

In the past I discussed about using the JuMP Package for solving an optimization problem. Recently I came across another problem that seemed like a good fit for the JuMP Package - The N-Queens puzzle.

Summary of the N-Queens Puzzle

So what is the N-Queens puzzle?

Imagine you have a chessboard and eight queens with you. The challenge is to place the queens in such a way that they do not attack each other with their usual moves.

Solution of the N-Queens Puzzle

This code listing is an attempt at solving the puzzle using the JuMP package from Julia:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
using JuMP, GLPK, Gadfly
N=8
model = Model(GLPK.Optimizer);
@variable(model, board[1:N,1:N], binary=true)
@objective(model, Max, sum( board[i,j] for i in 1:N, j in 1:N))

for j in 1:N
    @constraint(model, sum(board[i,j] for i in 1:N) == 1)
end
for i in 1:N
    @constraint(model, sum(board[i,j] for j in 1:N) == 1)
end

for k in 2:2*N
    @constraint(model, 
    sum(board[i,j] for i in 1:N, j in 1:N if i+j == k) <= 1)
end

for k in -(N-1):(N-1)
    @constraint(model, 
    sum(board[i,j] for i in 1:N, j in 1:N if i-j == k) <= 1)
end

print(model)
optimize!(model)
spy(JuMP.value.(board))
  • Line 1 is about importing the required packages. We’re using JuMP for formulating our linear programming problem, GLPK is one of the solvers available in Julia. Gadfly is used to visualize our result.

  • In line 2 we parameterize our Chess board.

  • In line 3 we define our model. We are asking JuMP to create a model with the GLPK optimizer.

  • In line 4 we model our Chessboard as a grid of an 8x8 matrix. Each cell in the matrix represents a square on the chess board. The cell has a value of 1 or 0 indicating whether the queen should be placed in the square or not.

  • In lines 7-9 we specify our first constraint. What we are saying here is that no two queens should be placed in the same column. This is because queens in the same column can attack each other.

  • In lines 10-12 we specify our second constraint. What we are saying here is that no two queens should be placed in the same row.

  • In lines 14-22 we are saying that no two queens should be in the same diagonal.

  • In line 25, we set the solver in motion using the optimize! command.

  • In line 26, we visualize the result using the BinaryMatrix visualization command spy from the Gadfly package.

What does Solving an Optimization Problem like N-Queens Entail?

N-Queens puzzle can be formulated as a linear programming problem. In this class of problems we model the puzzle and constraints as set of linear expressions. There are other ways of solving this problem, for example, using brute force method. Tere are many advantages of expressing the problem in an abstract linear program.

There are two styles of programming - Procedural and Declarative. In procedural programs we specify the path that the solver should take to find the solution to the program. We specify a series of steps and the computer takes them in the specified order. In declarative programs, on the other hand, the problem is specified in a domain-specific language and a solver or planner does all the hard work of finding a solution.

When you write a Java program to find the root of an equation using Newton-Raphson or other method, you are using a procedural style of programming.

When you are querying a relational database using an SQL-like language, you are using a declarative style of programming.

Two thing stand out when you use a declarative style of programming.

  1. Knowledge of the domain-specific language

    When you use declarative style, you should know the DSL. For example, when you query a database, you should be able to formulate a query.

  2. You should be able to translate your problem into the language of the domain

    Modeling is the name of the game, and over the course of time I have realized that it is more of an art than an exact science.

Another convenience that Julia offers is that we don’t have to understand the underlying solver which solves the Linear Programming problem. The solvers use an algorithm like the Simplex Algorithm to solve these problems.

A similar problem - The Confused Queens

I came across a similar puzzle which is an antithesis of the original N-Queens puzzle - The confused Queens puzzle, or what I would like to call as “Everybody was Chess-fu fighting”. Here we have to find an arrangement where all the Queens do attack each other. That’s for later!