How can I clear specific Python classes/modules from memory without using "clear classes"

I have successfully developed a workflow that will allow me to call custom python code from Matlab, see the following two scripts:

MATLAB:

function out = reloadPy() % A Matlab function
clear classes
mod = py.importlib.import_module('mymod');
py.importlib.reload(mod);
v1 = 1:5;
v2 = 6:10;
out = double(py.mymod.add2Vectors(v1,v2));

Python:

# mymod.py - a python function

import numpy
import array

def add2Vectors(v1, v2):
    myarray = numpy.array(v1) + numpy.array(v2)
    return array.array('d', myarray)

If I call reloadPy from the command line I get the following output:

>> test = reloadPy
Warning: Objects of 'onCleanup' class exist.  Cannot clear this class or any of its superclasses. 
> In reloadPy (line 2) 
Warning: Objects of 'table' class exist.  Cannot clear this class or any of its superclasses. 
> In reloadPy (line 2) 
Warning: Objects of 'datetime' class exist.  Cannot clear this class or any of its superclasses. 
> In reloadPy (line 2) 

test =

       7     9    11    13    15

>>

The warnings are stemming from the "clear classes" command. I am using that sequence of commands (lines 2-4) to reload the python module while debugging. Any changes I make to mymod.py are not captured unless I execute all 3 commands. The documentation for "clear" states:

Calling clear all, clear classes, and clear functions decreases code performance, and is usually unnecessary. ...

To clear a particular class, use clear myClass.

Is there a way to just clear just my python class or module? I can't find the name. I tried

[M, X, C] = inmem

But none of the result seem to be related to python.

[Emil Geiger]

Update: I can't use this workflow if I pass a variable into reloadPy because clear classes will delete the variable. I have also submitted a support ticket.

Answer 1

[Emil Geiger]

I received an answer from Support. Clearing just the python module is not possible at this time. But wrapping the reload statements in a function is a valid work around because any user variables are out of scope. For example:

function reloadPy()
warning('off','MATLAB:ClassInstanceExists')
clear classes
mod = py.importlib.import_module('mat');
py.importlib.reload(mod);

[Minchae Jeong]

Love you my brother this helped me a lot!

[Arwel]

Thanks Emil! This also worked for me, and this makes it general for my use-case (via a file)....

function reloadPy(name)

% Save the module name (so we can reload it after clear..)
save('tempSave','name');

warning('off','MATLAB:ClassInstanceExists')
clear classes

% Reload the name....
name = load('tempSave');
name = name.name;

% Now reload the module....
mod = py.importlib.import_module(name);
py.importlib.reload(mod);

% Get rid of our temporary file...
delete('tempSave.mat');

Answer 2

[Fan Yang]

Say, if we want to reload 'util.py':

py.util = py.importlib.reload(py.importlib.import_module('util'));

This line would work for me. It actually redefined a struct with name 'py', so it may overwrite other module you imported, but if all of your python module is explicitly imported in this way, there seems to be no problem. Anyway this should be regarded as a workaround as well.

[Peter Qiang]

This question has been bothering me for a long time. Many thanks for your solution！

Is a very useful method because in my 2021b with py 3.9, command line like:

clear classes
myModule = py.importlib.import_module('myModule');
py.importlib.reload(myModule);

just not working properly in my environment.

I guess cache still exist in somewhere after run that codes. So overwrite it seems become the only way works for me. To avoid overwrite matlab built-in py module I changed module variable name by remove perfix "py." , like:

myModule = py.importlib.reload(py.importlib.import_module('myModule'));

then call our module function with:

myModule.myFunction(args)

Thus it won't overwrite py. It works for me, at least, for now.

[Lucademicus]

It does not work for me in 2023B. Does anyone have another solution for me?

[Arwel]

Ditto for me - no joy with 2023b....