Try our new documentation site (beta).
Filter Content By
Version
Text Search
${sidebar_list_label} - Back
Filter by Language
sudoku.m
function sudoku(filename) % Copyright 2023, Gurobi Optimization, LLC */ % % Sudoku example. % % The Sudoku board is a 9x9 grid, which is further divided into a 3x3 grid % of 3x3 grids. Each cell in the grid must take a value from 0 to 9. % No two grid cells in the same row, column, or 3x3 subgrid may take the % same value. % % In the MIP formulation, binary variables x[i,j,v] indicate whether % cell <i,j> takes value 'v'. The constraints are as follows: % 1. Each cell must take exactly one value (sum_v x[i,j,v] = 1) % 2. Each value is used exactly once per row (sum_i x[i,j,v] = 1) % 3. Each value is used exactly once per column (sum_j x[i,j,v] = 1) % 4. Each value is used exactly once per 3x3 subgrid (sum_grid x[i,j,v] = 1) % % Input datasets for this example can be found in examples/data/sudoku*. % SUBDIM = 3; DIM = SUBDIM*SUBDIM; fileID = fopen(filename); if fileID == -1 fprintf('Could not read file %s, quit\n', filename); return; end board = repmat(-1, DIM, DIM); for i = 1:DIM s = fgets(fileID, 100); if length(s) <= DIM fprintf('Error: not enough board positions specified, quit\n'); return; end for j = 1:DIM if s(j) ~= '.' board(i, j) = str2double(s(j)); if board(i,j) < 1 || board(i,j) > DIM fprintf('Error: Unexpected character in Input line %d, quit\n', i); return; end end end end % Map X(i,j,k) into an index variable in the model nVars = DIM * DIM * DIM; % Build model model.vtype = repmat('B', nVars, 1); model.lb = zeros(nVars, 1); model.ub = ones(nVars, 1); for i = 1:DIM for j = 1:DIM for v = 1:DIM var = (i-1)*DIM*DIM + (j-1)*DIM + v; model.varnames{var} = sprintf('x[%d,%d,%d]', i, j, v); end end end % Create constraints: nRows = 4 * DIM * DIM; model.A = sparse(nRows, nVars); model.rhs = ones(nRows, 1); model.sense = repmat('=', nRows, 1); Row = 1; % Each cell gets a value */ for i = 1:DIM for j = 1:DIM for v = 1:DIM if board(i,j) == v model.lb((i-1)*DIM*DIM + (j-1)*DIM + v) = 1; end model.A(Row, (i-1)*DIM*DIM + (j-1)*DIM + v) = 1; end Row = Row + 1; end end % Each value must appear once in each row for v = 1:DIM for j = 1:DIM for i = 1:DIM model.A(Row, (i-1)*DIM*DIM + (j-1)*DIM + v) = 1; end Row = Row + 1; end end % Each value must appear once in each column for v = 1:DIM for i = 1:DIM for j = 1:DIM model.A(Row, (i-1)*DIM*DIM + (j-1)*DIM + v) = 1; end Row = Row + 1; end end % Each value must appear once in each subgrid for v = 1:DIM for ig = 0: SUBDIM-1 for jg = 0: SUBDIM-1 for i = ig*SUBDIM+1:(ig+1)*SUBDIM for j = jg*SUBDIM+1:(jg+1)*SUBDIM model.A(Row, (i-1)*DIM*DIM + (j-1)*DIM + v) = 1; end end Row = Row + 1; end end end % Save model gurobi_write(model, 'sudoku_m.lp'); % Optimize model params.logfile = 'sudoku_m.log'; result = gurobi(model, params); if strcmp(result.status, 'OPTIMAL') fprintf('Solution:\n'); for i = 1:DIM for j = 1:DIM for v = 1:DIM var = (i-1)*DIM*DIM + (j-1)*DIM + v; if result.x(var) > 0.99 fprintf('%d', v); end end end fprintf('\n'); end else fprintf('Problem was infeasible\n') end