I try to create a matrix (235 x 1200000) but I get the following error message : "Matrix -- Memory exhausted". I can create matrices until the size (100 x 1200000) only. I have 2Go of RAM. On another machine (which is not mine) the matrix is created normally although the RAM is only 1 Go.
What do you think the problem is due to ?
Thanks ---- Atidel
It's not a bug, and it has nothing to do with Lush:
On a 32-bit machine, you can't have arrays larger than 1GB or 2GB (depending
on how Linux was compiled). If you need bigger arrays, get a 64 bit machine
and install a 64-bit version of Linux.
Thank you Yann, but I am not very convinced by the answer.
Does the machine architecture (in particular the processors bits numbers) constraint the choice of dataset size that I choose for NN training ? knowing that NNs have been effectively used many decades ago when computers were not as performant as nowdays.
Is not the processor bits number but the capacity of an integer to address memory. Pointers to memory are interger or long type data, both 32 bits wide in a 32 bits linux. Thus you can address a maximum of 2^(32-1) memory words (normally bytes). That is 2147483648 bytes or 2Gb ... this limits several things in a system: theoretical program memory size, array size, file size, etc. and in older linux, partition sizes. With a linux 64 bits the limit is 2^(64 -1), a very very big number. Try it !