Is there a command to compress a matrix in memory in MATLAB?

I would like to be able to compress a matrix in memory, so that I can do calculations on larger matrices.

Answer 1

[MathWorks Support Team]

The ability to compress a matrix that is stored in memory is not available in MATLAB. The numerical data that is usually stored in the matrices is already stored in an efficient format.

A compression algorithm thus would not achieve any meaningful compression rates unless a lot of the elements are the equal. Compression algorithms are very time consuming and would slow down calculations.

To work around this issue, if your matrix contains a large amount of elements that are equal to cero, use the function SPARSE to create a sparse matrix that will store the matrix in a more memory-efficient way. Furthermore, and depending on your application, using a data type will less precision will also reduce the memory requirements of a large matrix.

[dpb]

andrei Answer moved to Comment @andrei, please reserve Answers for actual solutions to posed Questions -- dpb

What if there are many equal elements or even sequences of elements?

Say if one would byte the time-performance hit of the compression itself ... is it possible?

[dpb]

It would still take having the original data to compress, wouldn't it?

[Steven Lord]

Depending on the structure of your data and what you want to do with it, there may be other alternatives. Some functions, like gmres or eigs, allow you to specify a function handle that computes some function of your matrix rather than providing the matrix itself. If you can perform the operations required to evaluate that function without explicitly creating the matrix, that may help you operate on larger data than you would otherwise be able to use.

Another option is to use a datastore to create a tall array to operate on data that cannot fit in memory.