Not enough computer memory to train the data...

If my LDA is to train the data but I have not enough computer memory to train the data, what should I do? I tried splitting the data and used one of it to train. However, how should I input the rest of the training data?

Answer 1

[per isakson]

• Putting the training data in a memmapfile might help.
• standard answer: memory is cheap

[Lester Lim]

What's a memmap file? Can matlab variables be saved into it? Pardon me newbie questions...

[per isakson]

You are welcome! The most important MATLAB command is doc. It's limitation is that you need to know a name. Try

    doc memmapfile

[Lester Lim]

I've read it, but still a bit unsure, I gather it is like not importing the data but rather, calling it from the excel without using memory is that right?

[Lester Lim]

Tried it, however when I tried to call it out from the GUI, it didn't work...

[per isakson]

• I don't know what you done!
• What error message you received.

However, I know

• it is possible to use a memmapfile-object like a huge matrix
• it is blasting fast - order of magnitude faster than reading from an excel file

[Lester Lim]

The orignial was to import an excel file and write it into a variable. Then I get down to work by separating the labels from the training data... It didn't work with my codes when I put in m = memmapfile(LDA data). Error was undefined variable...

[Walter Roberson]

Write the imported data to a file in binary format. You can then memory map the binary file.

[Lester Lim]

But by writing into a binary format, won't everything become ones and zeros?

[per isakson]

See Binary file

[Lester Lim]

Il try...