Pre-allocate memory to zeros or NaN?

Question

Lina Koronfel 2021-8-5

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/892647-pre-allocate-memory-to-zeros-or-nan

评论： Lina Koronfel 2021-8-5

I'm dealing with large matrix size in a for loop and I currently pre-allocate the matrix to zeros before the for loop, however, this causes me some problems with some math equations I must apply on the matrix. So my question is, does pre-allocating the memory to a NaN matrix recommended? or does MATLAB not like that?

Update for clarification:

Example:

a=zeros(5,5)
for j=1:4:5
    v=j;
    a(:,j)=v+10;
    end     

So the problem is if I want to do any calculation with the rows, I will end up including rows 2,3,4 which are now 0, this messes up with my calculations. So I want to explore the option of creating a NaN matrix and using 'omitnan' in my calculations.

2 个评论
显示无隐藏无

KSSV 2021-8-5

It depends on what numbers you are going to fill into matrix. If you fill with zeros what problem it is causing? First what for the loop is?

Lina Koronfel 2021-8-5

在 MATLAB Online 中打开

The numbers are double precision. The problem is that I'm skipping some trials, so here is a quick example:

a=zeros(5,5)
for j=1:4:5
    v=j;
    a(:,j)=v+10;
    end     

So the problem is if I want to do any calculation with the rows, I will end up including rows 2,3,4 which are now 0, this messes up with my calculations. I know there is an option of skiiping every forth row but I don't want to do that because The code is pretty big and I would have to go to every line and change it. So I want to explore the option of creating a NaN matrix and using 'omitnan' in my calculations.

Please let me know if this is recommended, and if there is any harm if using NaN for pre-allocation. Thank you

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Chunru 2021-8-5

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/892647-pre-allocate-memory-to-zeros-or-nan#answer_761247

编辑：Chunru 2021-8-5

If you have skipped some data and want to omit those data later on, then the preallocation with nan is a better options. Especially if you want to use some matlab built in functions (such as mean, std), options are available to omit nans. NaN is the part of IEEE arithmetic representation, it should not have side effect on preallocatio. The only side-effect is when performing calculation (NaN+2=Nan), but you are prepared to ignore NaNs.