How to group rows of data into a cell of a cell array?

I have a data that has xyz information as well as three numbers that group each point together. As an example a row might have [ -1 2 -4 1 2 15] So that atom is at point (-1,2,-4) and it is defined by 1,2,and 15. There is an unknown amount of atoms that have different locations but have the same defined numbers. I want to group all of the atoms that have the same 3 defining numbers in a matrix that is in a cell array (or into structure if that is easier) QN. I have the following code but I am not adding more data to one cell. This is what I have

j=1;
for k = 1:(length(TypesQN)-1)
    for p = 1:(length(locQN)-1)
        if TypesQN(k,1:4)==locQN(p,4:7)
            qn=locQN(p,:);
            QN{j}=qn;
            j=j+1;
        end
    end
end

Where the amount of defined numbers is know (TypesQN) and the total dataset length is known (locQN).

Answer 1

[Cedric]

Here is one way. We start by building a fake data set:

 >> A = [rand(10,3), randi(2,10,3)]
 A =
    0.7513    0.8407    0.3517    1.0000    1.0000    1.0000
    0.2551    0.2543    0.8308    1.0000    2.0000    1.0000
    0.5060    0.8143    0.5853    2.0000    1.0000    1.0000
    0.6991    0.2435    0.5497    2.0000    2.0000    2.0000
    0.8909    0.9293    0.9172    2.0000    1.0000    1.0000
    0.9593    0.3500    0.2858    1.0000    2.0000    2.0000
    0.5472    0.1966    0.7572    2.0000    1.0000    2.0000
    0.1386    0.2511    0.7537    1.0000    2.0000    2.0000
    0.1493    0.6160    0.3804    1.0000    2.0000    1.0000
    0.2575    0.4733    0.5678    1.0000    2.0000    1.0000

Then we find group IDs per unique combination of 3 "ID" numbers:

 >> [~,~,gId] = unique( A(:,4:end), 'rows' )
 gId =
     1
     2
     4
     6
     4
     3
     5
     3
     2
     2

and we use these to split the array:

 >> groups = splitapply( @(x){x}, A, gId )
 groups =
  6×1 cell array
    {1×6 double}
    {3×6 double}
    {2×6 double}
    {2×6 double}
    {1×6 double}
    {1×6 double}

where you can see that the 3 numbers are the same for e.g. group 2:

 >> groups{2}
 ans =
    0.2551    0.2543    0.8308    1.0000    2.0000    1.0000
    0.1493    0.6160    0.3804    1.0000    2.0000    1.0000
    0.2575    0.4733    0.5678    1.0000    2.0000    1.0000

Note that if you don't want to save the ID numbers in the groups, you can split as follows:

 >> groups = splitapply( @(x){x(:,1:3)}, A, gId )
 groups =
  6×1 cell array
    {1×3 double}
    {3×3 double}
    {2×3 double}
    {2×3 double}
    {1×3 double}
    {1×3 double}

 >> groups{2}
 ans =
    0.2551    0.2543    0.8308
    0.1493    0.6160    0.3804
    0.2575    0.4733    0.5678

[Federico Lopez]

Hi Cedric, I have a very similiar need for applying this solution, but instead of having "A = 10x6 double", I have: "A = 65x12 table" . I already divided into groups with unique, and got the "gId". The step that I can't seem to get right is the splitapply.

How can I adapt the splitapply func to my needs? the @(x){x} does not work for me.

Please help!! Thank you very much

[xinyuan wu]

splitapply fcn can only be used in numeric class. I'm also looking for a substitute method. and now I'm using a fcn made by myself

[xinyuan wu]

@Federico Lopez

[Walter Roberson]

@xinyuan wu

splitapply can be used for cell arrays as well, and it is not limited to doing arithmetic computations on the fields. For example, the following gets the T.Gender fields and for each one takes length() [which would happen to be consistent for each gender]

What datatype do you need to work with? What non-numeric fields do you need to apply functions to?

load patients
T = table(Gender,Height);
[G,gender] = findgroups(T.Gender);
splitapply(@(x) {x{1}, cellfun(@length,x.')}, T.Gender, G)

ans = 2×2 cell array
    {'Female'}    {[6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6]}
    {'Male'  }    {[            4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4]}

Answer 2

[Walter Roberson]

TypesQN = [randi([-5 5],1000,3), randi(5,1000,3)];    %sample data for testing

[~, ~, group_number] = unique(TypesQN(:,4:6), 'rows');
grouped_atoms = splitapply(@(M) {M}, TypesQN, group_number);

The above requires R2015b or later; if you have an earlier version then the splitapply can be rewritten to other code.