Find the rows in matrix B that contain the two numbers in each row of matrix A, in every possible order

I have a Nx2-matrix A, and a Mx3-matrix B, with M>N. I would like to find the rows in B that contain the two numbers in each row of matrix A, in every possible order.

Example:

A = [1 2,
    2 5]

B = [1 5 6,
     1 4 2,
     2 9 3,
     9 5 2]

Then, for the first row in A, the second row in B should be selected, and for the second row in A, the fourth one should be selected. Thus, the result should be:

C = [1 4 2, 
     9 5 2]

I have already solved it with for loops, find and ismember, but my matrices are quite big and it is a quite time consuming solution. Can you please help me?

[Geoff Hayes]

Paula - please provide a sample of the code that you are using to determine the matrix C so that we can offer alternative suggestions. And also clarify what you mean by your matrices are quite big. What is the smallest and what is the largest matrix that you are concerned with?

Answer 1

[Mohammad Abouali]

have you considered parallel processing?

You could start MATLAB parallel workers using:

parpool % if you don't it would usually start it automatically.

It takes some time for parpool to start.

Now defining your Matrices (these are too small for parallel processing, so the cost of starting parallel workers is much more than doing the processing in parallel. So use your own big matrices here. A and B are broadcast variables. so they are communicated to all workers. the cost of that communication still might not justify it. so you really need to test it on your machine and matrices to see if parallel processing solves anything here.)

A = [1 2; ...
     2 5];
B = [1 5 6; ...
     1 4 2; ...
     2 9 3; ...
     9 5 2];

Now the parallel code that checks rows of B and A

nRa=size(A,1);
nRb=size(B,1);
results=zeros(nRa*nRb,1);
parfor idx=1:(nRa*nRb)
  [rB,rA]=ind2sub([nRb,nRa],idx);
  results(idx)=all(ismember(A(rA,:),B(rB,:))); %#ok<PFBNS>
end
results=reshape(results,nRb,nRa);

Note that the columns in results refer to rows in A and the rows in results refers to rows in B. So if row 2 column 1 of results is set to true or 1, then it means that row 2 of matrix B contained all the members of row 1 of matrix A.

You could check all the rows as follow: (I don't know if the rows of A and B have a one to one match or it is possible that multiple rows of B contain all the elements of, let's say row 1, of matrix A.

for rA=1:nRa
  fprintf('The folowing rows of B where in row %d of A:\n',rA);
  fprintf('%d ',find(results(:,rA)));
  fprintf('\n');
end

The folowing rows of B where in row 1 of A:
2 
The folowing rows of B where in row 2 of A:
4

Now to form your matrix C do this:

[rows,cols]=find(results);
C=B(rows,:);

C =

       1     4     2
       9     5     2

[Paula Radel]

Awesome! Thanks a lot :)

[Mohammad Abouali]

you are welcome

[Oluropo Dairo]

Hi Mohammad,

I have a sample dataset of the form below and I want to sort into group of A = 11 - 20, B = 21 - 30, C = 31 - 40, etc. using the second column to do the sorting. I tried this

N = sortrows(N,2); if N(:,2)<=0; N0 = N; elseif (N(:,2)>=1) & (N(:,2)<=10); N10 = N; elseif (N(:,2)>10) & (N(:,2)<=20); N20 = N; elseif (N(:,2)>20) & (N(:,2)<=30); N30 = N; elseif (N(:,2)>30) & (N(:,2)<=40); N40 = N; if N(:,2) > 40 N50 = N50(N); end

but nothing was displayed

20.13 79.52 1007

20.15 79.65 1007

19.82 80.3 1007

18.68 83.2 1006

18.55 84.3 1006

18.2 85.5 1006

17.99 85.7 1006

17.85 86.1 1006

17.45 86.7 1006

16.16 90.3 1007

16.07 88.9 1007

15.91 89.1 1007

19.87 67.48 1008

20.45 61.71 1008

21.62 53.97 1009

22.02 47.95 1009

18.54 82.9 1006

18.66 82 1006

18.46 84.1 1006

18.36 83.6 1006

24.38 54.33 1008

25.18 50.33 1008

25.9 46.73 1008

26.46 45.79 1008

30.29 31.58 1006

30.33 29.83 1006

31.76 24.36 1004

32.37 24.63 1004

32.38 26.74 1004

32.23 24.22 1004