Parfor problem --- "Array construction from ByteBuffer Threw an exception"

Hi! I run my code using the 'parfor' function BUT receive an error message:

• Array construction from ByteBuffer Threw an exception.

I am wondering what does this message mean in simple language? Any simple example could be great!

Thanks for your help!

[Edric Ellis]

This is an unexpected internal error. Can you provide simple reproduction steps? (Or you could contact MathWorks support directly for help with this).

[Jiashun Wu]

Thanks for your comments.

My idea is quite simple, I want to separate table records into different groups (in cells) based on the value of a variable (e.g., "varname_a"), so that I can deal with different specfic records differently. As there are millions of rows with thousands of different variable values, I would like to use the parallel pool to speed up this approach.

Let me try to write this structure in a general format:

Tablename_Cell = {};
K = table(unique(Tablename.varname_a), ...
          'VariableNames', {'varname_a'});
A_number = 1000;

O = parpool('local', Num);

parfor i = 1:A_number
    
    Tablename_Cell{i, 1} = Tablename( ...
        find(Tablename.varname_a == K.varname_a(i)) , :);
    
    % Some other codes afterwards --- But even just keep the code
    %   above would receive the error message "Array construction
    %   from ByteBuffer Threw an exception."
    
end

[Walter Roberson]

Did you time findgroups()?

[Jiashun Wu]

@Walter Roberson Thanks for you note BUT not sure what do you mean...

"find" is the Matlab function find.

[Walter Roberson]

"I want to separate table records into different groups (in cells) based on the value of a variable (e.g., "varname_a"),"

G = findgroups(varname_a) ;
Tablename_cell = splitapply(@(x) {x}, Tablensme, G);

[Jiashun Wu]

@Walter Roberson Thanks again for your note.

My aim is to use 'parfor' to help separate my table into sub-groups, without imposing any summary statistics. In other words, somewhere like improving the code in https://www.mathworks.com/matlabcentral/answers/535864-how-do-you-split-a-table-into-sub-tables-based-on-entries-in-a-specific-column.

Anyway, I have already improved my code for better efficiency while keeping a tradtional 'for' approach.

Answer 1

[Alvaro]

The following code runs with no issues in my version of MATLAB R2022b.

load patients
% Create test table
Tablename = table(Gender(1:5), Height(1:5), 'VariableNames', {'Gender', 'Height'});
clearvars -except Tablename

Tablename_Cell = {};
K = table(unique(Tablename.Gender), ...
    'VariableNames', {'Gender'});
A_number = 2;

parfor i = 1:A_number
    
    Tablename_Cell{i} = Tablename(find(strcmp(Tablename.Gender, K.Gender(i))),:);
    
    % Some other codes afterwards --- But even just keep the code
    %   above would receive the error message "Array construction
    %   from ByteBuffer Threw an exception."
    
end

I did switch from using == to strcmp since I got the error

Operator '==' is not supported for operands of type 'cell'.

but I believe everything else is the same. Consider trying Bowei's suggestion of using a for loop with a dummy table to determine whether parfor is the cause of the issue.

I would also take another look at the documentation that you linked

https://www.mathworks.com/matlabcentral/answers/535864-how-do-you-split-a-table-into-sub-tables-based-on-entries-in-a-specific-column

Functions like splitapply already include parallelization as part of their extended capabilities.

https://www.mathworks.com/help/matlab/ref/splitapply.html#refsect-extended-capabilities

Answer 2

[Bowei Li]

Hi, I run into this error when using parfor as well. I found there was a undefined variable in my parfor loop. To check it, you may change your parfor to for, run it, and see what it says.