num2cell at high dimensions - is this a bug?

Matlab help of num2cell(A,[DIM1, DIM2, ...]) says

"All DIMn inputs must be an integer with a value from NDIMS(A) to 1."

The function though does accept DIMs larger than NDIMS(A), yet gives strange looking output. Here is an example:

>> cell2mat(num2cell([1 2],[2 4]))
ans =
     1     2
>> cell2mat(num2cell([1 2],[4 2]))
ans(:,:,1,1) =
     1
ans(:,:,1,2) =
     2

Is this sensible at all? Or, is this a bug?

Answer 1

[Kavya Vuriti]

Hi,

The function ‘num2cell’ is intended to work properly when all the values of dim vector are between 1 and ndims(A), if A is input array. When a value greater than ndims(A) is given, it works the following way:

The input array (say A) is of size 5 x 6 for example, which can be interpreted as 5 x 6 x 1. When dim is in order, each cell in the output cell array has a numeric array with the same size as A, except with size 1 in the dimensions NOT specified by dim. num2cell with different values of dim argument gives the following results:

c = num2cell(A,1): each cell in c is 5x1(x1)

c = num2cell(A,2): each cell in c is 1x6(x1)

c = num2cell(A,3): each cell in c is 1x1(x1)

c = num2cell(A,[1 2]): each cell in c is 5x6(x1)

c = num2cell(A,[1 3]): each cell in c is 5x1(x1)

c = num2cell(A,[2 3]): each cell in c is 1x6(x1)

This behavior of num2cell, when the values in dim vector are greater than ndims(A) is due to “size function returns value 1 when queried dimension is greater than ndims(A)”.

When dim is not in order, the result will be: “take the result for num2cell(A,sort(dim)), and then permute each cell's dimensions to get the right order of the dimensions specified by dim”. So, the output looks the following way:

c = num2cell(A,[2 1]): each cell in c is 6x5(x1)

c = num2cell(A,[3 1]): each cell in c is 1x1x5

c = num2cell(A,[3 2]): each cell in c is 1x1x6

The number of cells can however be computed as numel(A)/prod([size(A,dim(1)),...,size(A,dim(N))]), where N is the length of dim vector.