Collect double values of a matrix

Hi,

I have got a question. I have a matrix with three columns (X,Y,Z coordinates) and nearly 60000 rows. Now, I would like to compare the X and Z values of all 60000 rows. All rows whose X and Z values only occur once should be deleted, only the rows whose X and Z value occur twice or more should be regarded. Next, the rows with the equal X and Z value should be compared and the row with the higher Y value should be written into a new matrix.

I thought about using a loop for the search of the rows with equal X and Z values but the computing time is quite high, is there a better way? And how do I define the "do"?

for k = 1:end

for m = k+1:end

if A(k,1) == A(m,1) && A(k,3) ==A(m,3)

do something

end

end

end

I haven't thought about the loop for the search of the higher Y value, yet.

Yours,

Tobias

Answer 1

[Andrei Bobrov]

Let xyz - your array (n x 3);

[~,~,c] = unique(xyz(:,[1,3]),'rows','stable'); % c - serial numbers of rows with same values in first and last columns. 
d = accumarray(c,1); % d - the number of specific rows with the same first and last columns. Here first row of 'd' c = 1, nth row - c = n.
[lo,i] = ismember(c,find(d > 1)); % duplicate rows in xyz.
j = accumarray(i(lo),find(lo),[],@(x)x(max(xyz(x,2)) == xyz(x,2))); % index of the row with the maximum value in the second column.
out = xyz(j,:);

[Tobias Riewendt]

Hello Andrei,

can you tell me something about the meanings of the different lines, please?

[Andrei Bobrov]

Please read about unique, accumarray, ismember.

[Tobias Riewendt]

Ok, I will do it. Thanks for your answer.

Answer 2

[Martin C.]

hmm well to count occurences you could somehow use a hashmap

containers.Map in matlab. that would be efficient.

Answer 3

[Turlough Hughes]

Ok Following your comment I've edited my answer:

I would first find rows which have equal X and Z values.

c=1;
for ii=1:size(A,1)
    if A(ii,1)==A(ii,3)
        id_s(c)=ii;
        c=c+1;
    end
end
B=A(id_s,:);

Rows in A with equal X and Z value and are assigned to B.

To ensure a given X (and hence a given Z) of a row is repeated in another row I find the rows where this is not the case and delete them.

[~,ix] = unique(B(:,1)); % so B(ix,1) includes all values in B(:,1) without repeating any.  
Xtw=B(:,1);
Xtw(ix)=[]; % Xtw - includes all values in X that occurred two or more times.
[~,i1x]=setdiff(B(:,1),Xtw);  % Gets index for rows in B where the x value is not repeated elsewhere.
B(i1x,:)=[];

[Tobias Riewendt]

That is not exactly what I want. I try to give an example:

[X , Y , Z]:

[1 2 3

2 5 3

7 7 7

7 7 9

1 5 3]

The result should be a matrix of the rows with an equal X AND Z value. All other rows should be deleted. This means:

[1 2 3

1 5 3]