Split large matrix into new smaller matrices

Question

If I have a matrix like so:

x = rand(256,144160);

How can I split it into smaller matrices that are 256 x 901?

I tried the mat2cell(x,256,901) but I was getting this error:

Input arguments, D1 through D2, must sum to each dimension of the input matrix size, [256 144160].'

rayryeng rayryeng · Accepted Answer · 2016-08-04T20:24:24

That's because you're not using mat2cell properly. What you need to do is specify how you want to segment out each dimension. The first dimension is fine as you want each matrix to have 256 rows, but for the second dimension you need 144160 / 901 = 160 matrices with 901 columns for each matrix.

As such, you need to specify a vector of 160 values with 901 for each element:

y = mat2cell(x, 256, 901*ones(1,160));

This tells mat2cell that you want 901 matrices where each matrix is 160 columns and all have 256 rows. This brings into light what the error message is saying. It's telling you that the way you want to split up this matrix, each dimension you are splitting up must all add up to the size of the original matrix. The first dimension is set to 256, so that's very obvious that you want all matrices to have 256 rows. For the columns, you must have 160 matrices of 901 columns each which thus add up to 160 x 901 = 144160. You only specified 901 and thus MATLAB complains telling you that 901 != 144160.

The output of mat2cell creates a cell array of matrices where to access the k^th matrix, you would use y{k}. The curly braces are very important. However, if I can recommend something, I would not use mat2cell and you should perhaps use reshape and create a 3D matrix where each slice is 256 x 901:

y = reshape(x, 256, 901, []);

Using cell arrays is very inefficient as it is designed to be a general purpose container. If you are planning to do numerical analysis or if you want to access multiple matrices at a time, stick with pure numeric types. An added benefit to using reshape is avoiding the headache of figuring out that you need 160 matrices in total. With this, you can leave one of the dimensions empty (i.e. []) which tells the reshape command to automatically determine how to fill this dimension with the elements given in the matrix x. Notice that I've done this judiciously with the third dimension. In other words, it'll calculate 160 for you automatically for the third dimension. Here, y(:,:,k) gives you the k^th matrix and is the k^th slice of the 3D matrix y.

Split large matrix into new smaller matrices

1 Answers