Running a large array

Hi,

I am trying to load in an array (the array is 362x332x75x1032 (34.7GB)), which exceeds maximum array size preference (16.0GB). How do i go about loading it in.

It is a variable which is part of a .nc dataset

Answer 1

[Star Strider]

See the documentation section on Tall Arrays (introduced in R2016b).

I have rarely needed to use them, so I have little experience with them.

[Sophie]

They seem to not work with netCDF files

[Star Strider]

I was not aware that it was a CDF file. There is a set of functions to work with netCDF Files, and another set of functions to work with CDF Files. They may have options that would work. (I rarely use CDF files, so I do not have extensive experience with them and functions using them.)

Answer 2

[John D'Errico]

Memory is cheap. Get more memory.

I'm sorry, but if you want to work with big data, you will often need sufficient capabilities to handle that data. No matter what, working with huge arrays, lacking sufficient will be slow. So your next question will be, how can I make my code run faster. Again ... get sufficient memory. Or, solve smaller problems.

[Sophie]

This is for my masters dissertation, so unfortunantly upgrading my memory isn't an option

[Walter Roberson]

@Sophie

Did the professor assign the hardware and say "you must run it on this hardware" ?

If not, then:

• if it is your own hardware, then there is always the option of upgrading it
• if it is university hardware, then there is always negotiating with the university to obtain an upgrade

Answer 3

[Voss]

You can read and process the file in sections, one section at a time, in a loop, by specifying the start and count input arguments to ncread.

https://www.mathworks.com/help/matlab/ref/ncread.html#mw_c40440f3-d75e-421a-a505-30f047a4054a

https://www.mathworks.com/help/matlab/ref/ncread.html#mw_5d90db5a-9e5c-49a9-a145-a94ee23fd848

And/or specify the stride argument to read only every so many values instead of all of them.

https://www.mathworks.com/help/matlab/ref/ncread.html#mw_d9c1dcb3-b394-4d44-af97-8bbcb933ebde