How Can I Take Samples from a Matrix Randomly?

Hi Everyone

I want to take samples from a matrix randomly which is called as in my case"matrix". I just simplified the "matrix", actually size of "matrix" is 65500x3 and size of "train" is 2560x3. So I don't want to do this manually. My purpose is generate one seperate "test" matrix from "matrix" and that first two column of "test" matrix must be different from first two column of "train" matrix. To make this I wrote this code, but it is poor and it doesn't work properly. Any idea?

train=[2 1 3;1 2 3;3 1 2;2 2 1;1 1 2];
matrix=[1 2 3;2 1 3;3 1 2;2 2 2;1 1 1;1 2 4;2 2 2];
test=[];
adj=6;
[rows cols]=size(train);

for j=1:adj

      flag1=[0];
      flag2=[0];
      flag3=[1];

      while flag3 ~= 1
          temp=datasample(matrix,1);

          for i=1:rows

              if sum(temp(1,1:2) == train(i, 1:2))==2
                  flag1=[flag1 1];
              end
          end

          flag3=sum(flag1 || flag2 == 1);
      end

      test=vertcat(test,temp);
  end

test
train

[Azzi Abdelmalek]

I want to take samples into a matrix randomly, but this samples 1. and 2. columns values can't be identical with another test matrix rows 1. and 2. colmns values

What does that mean?

Answer 1

[Stephen23]

Try this:

>> idx = ~ismember(matrix(:,1:2),train(:,1:2),'rows')
>> test = matrix(idx,:)

It compares only the first two columns, and returns an index where the matrix data is not in train.

But with your example train data there are no rows that are not already matches for the first two columns, so here is an example with a simpler train matrix:

>> train = [2 1 3;1 2 3;3 1 2]
train =  
   2   1   3
   1   2   3
   3   1   2
>> idx = ~ismember(matrix(:,1:2),train(:,1:2),'rows');
>> test = matrix(idx,:)
test =
   2   2   2
   1   1   1
   2   2   2

if you only want the unique rows then run this:

>> test = unique(matrix(idx,:),'rows')

[Aliyar Gunes]

Thank you Stephen, this is what I want. Thanks a lot.