Load Part of a Matrix or Cell array

Hey All,

I was wondering if the latest version of Matlab R2011a has any built in functionality, or if there's anything recent on the file exchange, that can load in only part of a vector, matrix or cell array stored in a ".mat" file.

For example if I have a file "output.mat" which contains a single variable "large_matrix" which is a (100x100) matrix is there a way I can load say the 10th element, large_matrix(10), without loading the entire matrix into memory?

The above is to just illustrate my question I know Matlab could easily load a 100x100 matrix into memory.

The most Matlab seems to offer is selectively loading variables e.g.

loaded_file = load('output','other_variable');
other_variable = loaded_file.other_variable;

But I'm hoping I'm wrong.

Any advice would be greatly appreciated, Thanks

Answer 1

[Bjoern Hendrik Fock]

Hello All,

Loading parts of arrays from .mat files have been discussed in the forum before:

http://www.mathworks.com/matlabcentral/newsreader/view_thread/11819 http://www.mathworks.com/matlabcentral/newsreader/view_thread/124558 http://www.mathworks.com/matlabcentral/newsreader/view_thread/171729 http://www.mathworks.com/matlabcentral/newsreader/view_thread/261264 http://www.mathworks.com/matlabcentral/newsreader/view_thread/277386 http://www.mathworks.com/matlabcentral/newsreader/view_thread/288926 http://www.mathworks.com/matlabcentral/newsreader/view_thread/301858

When I did an official feature request some years ago ( making-a-feature-request ) I got the answer that not enough users are requesting this feature. So if you like to have it, you should consider writing to MathWorks.

Best regards, Bjoern

[Jan]

+1. A nice summary of the frequency this feature is requested here.

Modern MAT-Files store the values of array in a compressed format. Accessing the values demands for a decompression of complete blocks. Therefore I assume, that it does not matter the efficiency if you cut out the wanted part inside the MAT reading function or in Matlab.

[Walter Roberson]

Jan, it could matter for memory use purposes. Suppose you wanted a small number of elements from a very large matrix. The decompression routine could read values and throw them away until it got to the correct place -- and if the values happened to be consecutive in the file, it wouldn't even have to hunt around and decompress other blocks.

There are, of course, efficiency issues such as if the user's request did not index the elements in monotonically increasing order and the list of requested elements was too large to sort in the available memory. None the less, there are cases where it would make sense for memory use and file efficiency to read selectively.

[Dyuman Joshi]

@Bjoern Hendrik Fock, the links you have attached don't seem to be working anymore.

Answer 2

[Titus Edelhofer]

Hi,

unfortunately the answer is no (for R2011a). What I have done is to write a simple function that splits the variable into pieces when saving to a file, and read the pieces afterwards while processing the data...

Titus

Answer 3

[Bjoern Hendrik Fock]

The feature seems to be added in Matlab R2011b by a new function called matfile. I will try this out the next days. MathWorks, this looks like an important improvement. Thank you for listening to your customers.

Bjoern