Non integer results when generating sparse matrix indicies

Hello,

I am using a large integer data series to generate subscript indices for a sparse matrix using:

clear all
close all
data = randi(100,[2,10000000]); % Example data
[Nb,Nt] = size(data); 
indx = 3*(data(:)'-1) + (1:3)';
Findx = reshape(indx,3*Nb,Nt); % An intermediate index used elsewhere
% sparse matrix indicies
Hindx1 = repmat(permute(Findx,[1 3 2]),1,3*Nb,1);
Hindx2 = permute(Hindx1,[2 1 3]);
% Are the integer?
find(Findx~=round(Findx))
find(Hindx1~=round(Hindx1))
find(Hindx2~=round(Hindx2))

The Findx is normally clear but the Hindx don't conform, usually there are only one or two values that are non-integer.

Is this a bug or is it a problem with my machine:

MATLAB Version: 9.7.0.1247435 (R2019b) Update 2

MATLAB License Number: 345666

Operating System: Linux 5.3.0-51-generic #44~18.04.2-Ubuntu SMP Thu Apr 23 14:27:18 UTC 2020 x86_64

Java Version: Java 1.8.0_202-b08 with Oracle Corporation Java HotSpot(TM) 64-Bit Server VM mixed mode

Thanks,

Mat

[Ameer Hamza]

I tried it on R2020a and R2018b, and this issue didn't happen. All the find() calls return an empty vector. Can you check how much difference occurs between the non-integer values and the corresponding integer values?

[Mat]

It varies but mostly the difference is 0.25 or 0.5. Also I have updated to 2019b v5 but the issue persists.

[Rashed Mohammed]

Hi Mat

I tried to reproduce the issue with your code on both 2019b Update 2 & Update 5 but I didn’t encounter the issue you mentioned. Try running the code on another machine or do a clean install of MATLAB to see if the issue occurs again.

[Mat]

Hello Rashed,

I think it may have been related to some sort of memory llimit, my machine has 16gb RAM and If I increased the size of the arrays much larger it would crash matlab entirely. I have moved to 2020 now and all seems fine. Also being able to use unsigned integers as indicies for sparse is a nice feature. Thanks for responding, I may try a fresh install of 2019b if I have the time.