File Exchange

## TSPSEARCH

version 1.1 (20.1 KB) by

Heuristic method for the Traveling Salesman Problem (TSP)

4.66667
6 Ratings

Updated

A number of nearest neighbour tours are generated from randomly selected starting points. Each tour is improved by 2-opt heuristics (pairwise exchange of edges) and the best result is selected.

zhu jinxiang

Michael Schroer

### Michael Schroer (view profile)

It's amazing how much more efficient this code is than the other algorithms on the File exchange I've seen. Very nice piece of code.

Also, for the lucky among us who have access to the parallel computing toolbox, a few small changes to the code allow the use of parfor in the bottleneck loop:

% par for loop
Lvals = inf*ones(m,1);
pvals = zeros(m,n);
parfor k = 1:m
% Nearest neighbour tour
p = greedy(s(k),D);
% Improve tour by 2-opt heuristics
[p,L] = exchange2(p,D);
% Keep best tour
Lvals(k) = L;
pvals(k,:) = p;
end
Lmin = min(Lvals);
pmin = pvals(Lvals==Lmin,:);

I don't fully understand how this gives me the speedup that it does, but I ran this for a particular problem of interest to me with 256 points. The standard code ran in 147 seconds, and the parfor loop (on my 12 core Xeon) ran in 0.87 seconds!

Mark

### Mark (view profile)

What is wrong with the sample code provided by Francesco? It works fine for me. He even gave you links to the mat files he was using.

abhishek

### abhishek (view profile)

Can anybody mail me exactly working program because whenever i try to run this code it shows some error like,
"Input argument "X" is undefined" and
"Maximum recursion limit of 500 reached. Use set(0,'RecursionLimit',N)
to change the limit. Be aware that exceeding your available stack space can

my mail id is abhishekdixit011@gmail.com
thanking you!!!

Mark

### Mark (view profile)

Seems to be an issue with the heuristic and my specific set of "cities". I tried it with uniformly distributed cities in a square area and it seems to behave well. I wish I could remove my previous rating....

Liber Eleutherios

### Liber Eleutherios (view profile)

Hi Mark, Jonas's is a heuristic method and it provides random results. But the speed of the algorithm and the efficiency of the path found are amazing compared to other alternatives currently submitted to File Exchange.
Try this simple example:

% Generate a 2D Random Walk with unitary step
N = 1000; % number of nodes
xy = randn(1, 2); % 2D coordinates of nodes
xy(N, 2) = 0;
for hh = 2:N
tmp = randn(2, 1);
tmp = tmp / sqrt(sum(tmp.^2));
xy(hh, 1) = xy(hh - 1, 1) + tmp(1);
xy(hh, 2) = xy(hh - 1, 2) + tmp(2);
end

% Calculate matrix of distances with function distmat from FE#15145
% http://www.mathworks.com/matlabcentral/fileexchange/15145-distance-matrix
distM = distmat(xy);

% Solve TSP problem with tspsearch
tic, [TSP_2OPT_P TSP_2OPT_C] = tspsearch(distM); toc
% Elapsed time is 9.769581 seconds.
% Show total distance / total cost
TSP_2OPT_C
% TSP_2OPT_C = 5.603308638831134e+002

% Have a look at the path: impressive!
figure
plot(xy(:, 1), xy(:, 2), '.b')
for hh = 1:N-1
hold on
plot(xy(TSP_2OPT_P(hh:hh+1), 1), xy(TSP_2OPT_P(hh:hh+1), 2), '-r')
pause(0.1)
end
plot(xy(TSP_2OPT_P([1 N]), 1), xy(TSP_2OPT_P([1 N]), 2), '-r')

% Now solve TSP problem using Yonathan Nativ's FE#24857 function - which is also very good:
tic, [xy_tsp ind_tsp] = solveTSP(xy, 0); toc
% Elapsed time is 9.771463 seconds.
min_cost = distM(ind_tsp(1), ind_tsp(end)) + sum(distM(sub2ind([N, N], ind_tsp(1:end-1), ind_tsp(2:end))))
% min_cost = 5.949210148822580e+002

% Have a look at the path: it's still very efficient! Only the last connection is unreasonable:
figure
plot(xy(:, 1), xy(:, 2), '.b')
for hh = 1:N-1
hold on
plot(xy(ind_tsp(hh:hh+1), 1), xy(ind_tsp(hh:hh+1), 2), '-r')
pause(0.1)
end
plot(xy(ind_tsp([1 N]), 1), xy(ind_tsp([1 N]), 2), '-r')

Mark

### Mark (view profile)

Is it just me, or does the algorithm seem to hang randomly once you go past ~350 points???

Liber Eleutherios

### Liber Eleutherios (view profile)

Best TSP engine on FE - much better than Yonathan Nativ's FE#24857 (which is now my second best).

Emmanuel Luevano

### Emmanuel Luevano (view profile)

Hey dude, It's a good job, easy to apply and very fast!!

 23 Feb 2012 1.1 Some examples added
##### MATLAB Release
MATLAB 7.11 (R2010b)