Grouping based on similarity and indices

I have an array A of size 300*1 which has a lot of repeating values. These values are the angular position of a point on a circle so these values range between (0,360). Is there a function in matlab that can group similar values present in array A and return the indices of these values. With these indices I can average data in another array B of similar size.

Answer 1

[John BG]

[v_sorted i_sort]=sort(v)

in v_sorted same angles are together, all values sorted in ascending order

i_sort contains the permuted indices so that

v_sorted=v(i_sort)

if you find my answer useful would you please mark it as Accepted Answer by clicking on the ACCEPT ANSWER button?

thanks in advance for time and attention

John BG

<mailto:[e-mail] [e-mail]>

[John BG]

thanks

[Jaime Castiblanques]

I have a similar problem. I have a matrix of 2904x3 where each column represents the x, y and z coordinates of some vectors. I need to group them in such a way that I can get several matrices of vectors which all have the same z coordinate. Any ideas?

[Image Analyst]

Use findgroups(). Post your data in a new question if you cannot figure it out.

Answer 2

[Sean de Wolski]

doc discretize

Then

doc accumarray
doc splitapply

Answer 3

[Image Analyst]

For example if you know how many groups there are, then you can use kmeans. Here's a full demo with 5 groups:

% Clean up / initialization
clc;    % Clear the command window.
close all;  % Close all figures (except those of imtool.)
clear;  % Erase all existing variables. Or clearvars if you want.
workspace;  % Make sure the workspace panel is showing.
format long g;
format compact;
fontSize = 20;

numGroups = 5
radius = 1
centerAngles = linspace(0, 330, numGroups+1)
allAngles = [];
subplot(1, 2, 1);
for g = 1 : numGroups
  theseAngles = [centerAngles(g) + 30 * rand(1, 10)]';
  allAngles = [allAngles; theseAngles];  
  % Plot
  x = cosd(theseAngles);
  y = sind(theseAngles);
  plot(x, y, '.', 'MarkerSize', 20);
  hold on
end
grid on;
axis square
title('Truth - Actual Groups', 'FontSize', fontSize);
xlim([-1.1, 1.1]);
ylim([-1.1, 1.1]);

% Now assume there are numGroups and figure out what group they belong to
% Cluster via kmeans()
groupLabels = kmeans(allAngles, numGroups);

% Plot
x = cosd(allAngles);
y = sind(allAngles);
subplot(1, 2, 2);
gscatter(x, y, groupLabels);
grid on;
axis square
title('Groups Determined by kmeans()', 'FontSize', fontSize);
xlim([-1.1, 1.1]);
ylim([-1.1, 1.1]);

The colors are different but that doesn't matter. Each group is identified by a different color. You can see that every point is grouped into the group that it should be classified into.

[Maria]

Hi, sorry if I comment on this old answer. I have just tried the code, but the groups' identification is wrong as you can see here

[Image Analyst]

Maria, that is definitely strange. I can reproduce it, but only for the very first run after opening MATLAB. Every other run works just fine - it's just the very first one that is bad. I don't know why. My guess is it has something to do with the random number generator seeding. I'll send it in to MATLAB and see what they say.

[Image Analyst]

Maria: below is the answer I got from the Mathworks. I am attaching the fixed version.

===============================================

I am writing in reference to your Technical Support Case #03085610 regarding 'kmeans grouping is not correct the first time'.

After further discussion with my peers, we were able to determine that the issue is arising due to the way in which "kmeans" calculates clusters. "kmeans" is an algorithm that is very sensitive to its initialization. For this particular problem it is very likely that the "kmeans" will be trapped in a local minima and cannot reach the global minima (i.e. the ground truth solution).

One approach to circumvent this issue is to repeat the clustering several times with different initializations by using the name value pair of ‘replicates’ within the "kmeans" function. Replace the line of the code where you call the "kmeans" function with the following:

>> groupLabels = kmeans(allAngles, numGroups, 'replicates',5);

[Maria]

Thank you very much! It's working now

Answer 4

[Image Analyst]

Yes, lots of them. Do you know how many clusters you will have? Have you checked out the Classification Learner app on the Apps tab? It requires the Statistics and Machine learning Toolbox. That's definitely your first place to start.