combine columns with different lengths to create a matrix

hi,

how can I create a matrix with vectors with different sizes as columns?

thanks,

[Guillaume]

You can't. By definition a matrix has the same number of elements in each row and the same number of elements in each column.

Which problem are you trying to solve?

[mehrdad asadi]

oh i know it's not possible by combining them without modification. u see, i mean is it possible to extend the vectors to a maximum size with zeros and then combine them?

[mehrdad asadi]

 for i = 1:sqrt(n)
    pci = find(any(cul==i)) %finding # of columns in each patch                                   that contains each column of the image
    for j = 1:max(size(pci))
        pc(:,:,j) = P_db(:,:,pci(j))
    end
end

u see, pci varies in each loop, this code does't work cause of matrix size mismatch. I want to modify 'pci's so that I can create 'pc'!

[Ashmil Mohammed]

a=rand(10,1); % randomly make two vectors replace this with your code
b=rand(8,1);
s=size(b);z=size(a);% (find size of both)

d=[b;zeros(z(1)-s(1),1)]; % concatenate the smaller with zeros
c=[a,d]; % catenate the two vectors

Answer 1

[Jos (10584)]

Many years ago, I wrote the function PADCAT to accomplish this. You can find it on the Matlab File Exchange:

http://www.mathworks.com/matlabcentral/fileexchange/22909-padcat

[mehrdad asadi]

that's fine,

thank you

[Monica Lopez Hidalgo]

nice!

[NMarco]

After 6 years, I found this and, @Jos (10584), you saved my life, my time, my code. Thanks!

Answer 2

[Jan Siegmund]

If the input data is a cell array of column vectors, you might consider this:

a   = {ones(5,1) ones(8,1)}; %test data
len = max(cellfun('length',a));
b   = cellfun(@(x)[x;zeros(len-size(x,1),1)],a,'UniformOutput',false);
out = [b{:}]

Answer 3

[Guillaume]

First, some comments on the code you've posted:

   for i = 1:sqrt(n)
      pci = find(any(cul==i)) 
      for j = 1:max(size(pci))
          pc(:,:,j) = P_db(:,:,pci(j))
      end
  end

You're using find to convert a logical array (returned by any) into an array of indices, and then using these indices to index into a matrix. The find is unnecessary work. Just use the logical indices directly.

Secondly max(size(v)) to find the numbers of elements in a vector is a very strange construct. numel(v) is the function you should be using.

Third, you don't need a loop to copy the elements. As a result, the code you've posted could be rewritten as:

for i = 1:sqrt(n)
   pci = any(cul == i); %pci a logical array
   pc = P_db(:, :, pci);
end

I'll also note that in the above code and in your example code the result pc is overwritten in each loop, so it does not matter if it changes size. I assume you meant to store it in something else. Finally, you mention concatenating column vectors, but according to your code pc is a 3D matrix.

Now, to answer your question, the best thing is to store pc in a cell array. Then it does not matter if it changes size on each iteration:

for i = 1:sqrt(n)
   pci = any(cul == i); %pci a logical array
   pc{i} = P_db(:, :, pci); %store in a cell array
end

Afterward, you could pad all these 3d matrices to the maximum size and concatenate them all as a 4D matrix, but I'm not sure why you'd want to:

allpcsizes = cellfun(@size, pci, 'UniformOutput', false);
max3dsize = max(vertcat(allpcsizes{:}));
for i = 1:numel(pc)
   resizedpc = zeros(max3dsize);
   resizedpc(1:size(pc{i}, 1), 1:size(pc{i}, 2), 1:size(pc{i}, 3)) = pc{i};
   pc{i} = resizedpc;
end
pc4d = cat(4, pc{:});

[mehrdad asadi]

actually your answer was really helpful. I used 'find' because I need the column indices with contain 'i'. without 'find' I get a vector with elements of 0 or 1 which I should again check for the indie of column.

 for i = 1:sqrt(n)
   pci = find(any(cul == i)); %pci a logical array
   pc{i} = P_db(:, :, pci); %store in a cell array
 end

I changed my code near to what you said and that's much better because using a cell array will help me to omit a loop.

but I should say that P_db is kind of a data-base. let 'pci' have 3 elements. so I'm storing 3 p*q matrices in one of the cells. how could I access for example the 2nd matrix in that cell. Do I have to use matrix instead of cell array?

finally, I appreciate your answer, that was a complete one!

[mehrdad asadi]

another thing is that after I create 'pc', I need to kind of average the matrices is each cell, how this is possible?

[Stephen23]

A few points:

• For the given code find is not required, and using logical indexing is neater and faster anyway.
• To calculate the mean try using the function mean. The second optional argument might be useful for you, so read the documentation carefully. Of course this would have to be applied to each cell of the cell array, for which a loop might be easiest.
• "how could I access for example the 2nd matrix in that cell" does not make any sense: one cell of a cell array can only contain one variable (which may be numeric, character, cell, etc), so it is not possible to have a second matrix in one cell. If however you are talking about a cell array, then just use cell indexing: pc{2} obtains the variable within the second cell of the cell array pc.

[Guillaume]

@mehrdad: . I used 'find' because I need the column indices with contain 'i'. without 'find' I get a vector with elements of 0 or 1 which I should again check for the indie of column

You don't need to use find, you can pass the vector of 0 and 1, a logical array as indices, matlab will only keep the elements for which the logical array is 1.

To access your '2nd matrix in a cell' (the wording is a bit sloppy, it's the 2nd page of the single matrix in the cell you want:

ma= pd{i}(:, :, 2); %2nd matrix in cell i

[mehrdad asadi]

yeah I got it,

Thanks a lot.