passing a mat file variable to a function

I have written a function myfun.m which works on a very large input variable var stored in a file myfile.mat in the current folder. If I first load var into the workspace and then call myfun(var), var requires memory of twice its size because it is kept both in the base and the function workspace. How can I instead pass just the names of myfile.mat and var on calling myfun, and load var from within myfun? Thank you!

Answer 1

[dpb]

A) Just put the call to load the variable in your function, perhaps with a call to uigetfile to enable the user to select the file interactively, or

B) Put the filename in the argument list instead of var and return var (or whatever is the desired result instead of having it in the argument list --

function var_perhaps_modified=myfun(inputfilename)
  ....

or,

C) Use nested functions and then the copy of var can be shared.

[Yaser Khojah]

Can you show please how the function var_perhaps_modified=myfun(inputfilename) works?

[dpb]

It's just a prototype for whatever the OP has that would accept the filename as a string variable instead of passing the array.

In the end, however, memory usage will probably be the same either way; ML will only "copy on write" if the argument is modified and if the resulting array is modified inside the function there will need be a copy for the target, anyways...

Answer 2

[Bernhard]

Hi dpb,

thank you for answering to my question.

The problem with just loading var from within myfun is that var needs not to have a unique name in myfun.mat.

My idea is that the user on calling myfun either passes a variable var directly from the workspace, or the name of a variable to be loaded from a matfile, together with the name of this matfile. Starting from your suggestion I have now found a way to do this using a cell array of strings {'myfile.mat','varname'}:

function varout = myfun(varin)
if iscell(varin)
    if length(varin) == 2 && ischar(varin{1}) && ischar(varin{2})
        filename = varin{1};
        varname = varin{2};
        load(filename,varname)
        var = eval(varname);
        clear(varname)
    else
        error('wrong format of input argument')
    end
else
    var = varin;
end
varout = do_some_calculations(var);
end

Renaming and then clearing the loaded copy of var ensures that at the time of the calculations only one copy of var resides in the memory. This is certainly not the most elegant way to do this but it works.

Bernhard

[Stephen23]

A much better solution would be to load into an output variable and then simply access its field:

S = load(filename,varname);
var = S.(varname);

https://www.mathworks.com/matlabcentral/answers/304528-tutorial-why-variables-should-not-be-named-dynamically-eval