clustering based on a custom function

I have set of nodes of which some are randomly selected to form a solution (or particle), and are known as 'clusterheads'. For example, I generate 20 random coordinates for nodes and pick 5 of these nodes as clusterheads. I form 6 Solutions so there will be 6 random selections of the original 20 nodes.

For each solution, I want the remaining nodes to cluster with a clusterhead based on a specific weight function. However when I run my code instead of getting the remaining nodes to cluster with the clusterhead to form an array for each clusterhead, I get arrays which are 5-10x the size of the original set of nodes. I find that there are many repeated values from these arrays.

For example, instead of getting 15 nodes connecting to various clusterheads, i get 75-150(number not consistent) 'nodes' connecting to the clusterheads. Each cluster has many repeated values. Any help would be appreciated.

@the cyclist

clc;   
clear;
close all;
%paramenters
nodes = 20;    % number of nodes
chead = 5;     % number of clusterheads for each solution
numparticles = 6;   % number of solutions/particles
maxrange = 50
%initialization
x = [rand(nodes,2)*100;50 50]; % set of random co-ordinates (or nodes) on 100x100 grid and a base station at (50,50)
ResEn = 20*rand(nodes,1);      % energy of each node (0-20J)
k = randperm(nodes);
emptyparticle.position = [];  % position of 5 clusterheads that i want all others to cluster to based on a function
emptyparticle.velocity = [];  % velocity of each solution partice 
emptyparticle.energy = [];    %energy of each solution particle
emptyparticle.cluster = cell(chead,1); %empty arrays that i want nodes to go to based on function
emptyparticle.cost = [];      %cost function based on clusterhead position and clusters, ignore for now
emptyparticle.best.position = []; % best position of each particle,ignore for now
emptyparticle.best.cost = [];  %best cost of each particle, ignore for now
emptyparticle.clustervalues = []; %an array that i created that stores the values of the cluster function
particle = repmat(emptyparticle,20,1);
for i = 1:numparticles                  %10 particles(or solution), so 10 solutions with 5 clusterheads each and each clusterhead has its own cluster of nodes
    k = randperm(nodes);
    emptyparticle.position.cluster = x(k(1:chead),:); %each solution is assigned a random set of nodes as cluster heads, for example particle(1).position would be [x1 y1;x2 y2;x3 y3;x4 y4;x5 y5]
    particle(i).position = x(k(1:chead),:);
    particle(i).velocity = 0;
    particle(i).energy = ResEn(k(1:chead),:);
end
%clustering process. For node j and clusterhead k, 
%a weighting function =
%Energy(clusterhead(k))/(dist(node(j),clusterhead(k))*dist(basestation,clusterhead(k))*nodedegree(clusterhead(k)))
%nodedegree is number of nodes already connected to clusterhead
for i = 1:numparticles
    for k = 1:chead
        for j = 1:nodes
            if pdist2(particle(i).position(k),x(j))>50 
% if distance between a ch and node is above range, weight value is 0
                 particle(i).clustervalues(k,j) = 0;
            elseif pdist2(particle(i).position(k),x(j))<50
                if size(particle(i).cluster{k},1) == 0 
%if cluster for clusterhead of a clusterhead is empty
                    particle(i).clustervalues(k,j) = ResEn(j)/(pdist2(particle(i).position(k),x(j)*pdist2(x(nodes+1),particle(i).position(k))));
                elseif size(particle(i).cluster{k},1) > 0 
%if cluster for clusterhead has a node/nodes in it
                    particle(i).clustervalues(k,j) = ResEn(j)/(pdist2(particle(i).position(k),x(j)*pdist2(x(nodes+1),particle(i).position(k))*size(particle(i).cluster{k},1)));
                end
                [m, ind] = max(particle(i).clustervalues(:,j)); 
%from the array created i want to find the maximum weightfunction for each node and cluster the node with the corresponding clusterhead
                particle(i).cluster{ind} = [particle(i).cluster{ind};x(j,:)];
            end
        end
    end
end

[Image Analyst]

Are you doing your own clustering? Why not use a built-in one like kmeans, knnsearch, dbscan, fitctree, etc.?

[kamran waheed]

uni work

Answer 1

[the cyclist]

I can't say that I have spent enough time with your code to fully understand it.

But, I notice the following. Inside the loop over k, you assign the clustervalues:

particle(i).clustervalues(k,j)

Then, with this line of code

[m, ind] = max(particle(i).clustervalues(:,j));

you are looking for the maximum clustervalue from the 5 clusterheads. However, that line of code is still inside the loop over k. So, I think you might inadvertently be taking that "maximum" multiple times:

• k=1: max of value 1
• k=2: max of values 1:2
• k=3: max of values 1:3
• etc

I think you want to pull that line of code outside the loop over k.

[kamran waheed]

Thanks for your advice and time, I really appreciate it and yes it works as intended now.