Putting similar numbers into groups within an array

I have an array of numbers that looks something like follows. I want to group the array into subgroups where the numbers are all within 2 of each other. In this case, there would be 3 groups. Is there an easy way to do this? I need the method to be automated for work for arrays 100 entries long.

7340.1
7340.3
7340.6
7349.0
7349.4
7358.0
7358.1
7358.2
7358.7
% New groups would look like follows:
% Group 1
7340.1
7340.3
7340.6
% Group 2
7349.0
7349.4
% Group 3
7358.0
7358.1
7358.2
7358.7

Answer 1

[Image Analyst]

If you have the Statistics and Machine Learning Toolbox (for pdist2) and the Image Processing Toolbox (for bwlabel), you can do this:

m = [...
7340.1
7340.3
7340.6
7349.0
7349.4
7358.0
7358.1
7358.2
7358.7]
% % New groups would look like follows:
% % Group 1
% 7340.1
% 7340.3
% 7340.6
% % Group 2
% 7349.0
% 7349.4
% % Group 3
% 7358.0
% 7358.1
% 7358.2
% 7358.7]
% First sort m so that close by ones has adjacent indexes.
m = sort(m, 'ascend')
% Get distance of every element to every other element.
distances = pdist2(m, m)
% Find out which pairs are within 2 of each other.
within2 = distances > 0 & distances < 2
% Erase upper triangle to get rid of redundancy
numElements = numel(m);
t = logical(triu(ones(numElements, numElements), 0))
within2(t) = 0
% Label each group with an ID number.
[labeledGroups, numGroups] = bwlabel(within2)
% Put each group into a cell array
for k = 1 : numGroups
  [rows, columns] = find(labeledGroups == k);
  indexes = unique([rows, columns]);
  groups{k} = m(indexes);
end
celldisp(groups); % Display the results in the command window.

It gives you your desired result. Should work for other arrays also, though I didn't test it with any others.

[Kyle Reagan]

This was extremely helpful! I'm now going to try to use it on large arrays of raw data. Thanks for the link to DBSCAN and the helpful code!

[Ayca Altay]

This answer does not work when we have groups of one elements. For example if data is as [1 10 12 13 14], 1 is a group by itself and the bwlabel command only allows groups with more than one element; so, 1 is lost in the process. How can we change that?

[Albert Mamani]

Ayca Altay did you find a way to do it?

[Image Analyst]

Depends on how Ayca defines groups. You could also say 10 is a group by itself as are 12, 13, and 14. So why are there not 5 groups? You need to specify somehow what a "group" means to you, like elements are closer than 1.5 to adjacent elements in the vector, or whatever.

[Mr.Alb]

Hi,

I'm still facing the @Ayca Altay's issue when groups hold just one element. How can I modify the previous script? I have an array like this:

a = (4.17, 8.33, 12.5, 16.67, 20.83, 25, 2.085, 6.245, 10.415, 14.585, 18.745, 22.915, 0.005, 4.165, 8.335, 12.505, 16.665, 20.835 ,2.08, 2.08, 6.25, 10.42, 14.58, 18.75);

Here, there are 13 groups (but they may vary depending on how a is made up, this is just an example), but I find only 10 groups with the method before. In particular, I lose the first group (1 occurence at about 0) and the two last groups ( 1 occurence each at about 23 and 25).

How can I solve this?

Many, thx

[Image Analyst]

You could try kmeans() and have it automatically check a variety of k values for the "best" k (number of groups).

Answer 2

[Image Analyst]

Looks like it could be a job for dbscan https://en.wikipedia.org/wiki/DBSCAN

[Alamgir M S M]

This is my Data-

[705.7142857 705.7142857 173.4285714 84.71428571 232.5714286 232.5714286 114.2857143 55.14285714 25.57142857 74.85714286 35.42857143 15.71428571 5.857142857 5.857142857].

I want to group same data into 1 group. This is 14*1 matrix where it has 3 pairs(705.7142857 705.7142857, 232.5714286 232.5714286 & 5.857142857 5.857142857) of same value data. So, I want these 3 pairs into 3 groups and rest of them into other 8 groups.

Is there anybody who can help me to code this?

[Image Analyst]

Use unique() and setdiff(). On the remainder, use kmeans() to group into 8 groups. Should be easy. If you can't figure it out, let me know.