chrchang / eigensoft

Experimental EIGENSOFT performance improvements.

Home Page:http://www.hsph.harvard.edu/alkes-price/software/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

large samples

sk0427 opened this issue · comments

Hi

I am trying to PCA with smartpca for 200K sample , 100K SNPS with 256G memory .
However, the process does not finish (core dumped). I suspect memory shortage, but I do not have access to additional memories.

Is this problem is reasonable(compared with memory size)? and Are there any way to solve it?

Thanks,

Satoshi

You should use the 'fastmode' from EIGENSOFT 6 or later (or equivalently, plink 2.0's "--pca approx" function); that approximation algorithm is simultaneously faster and more memory-efficient in this case.

Thanks! But I already tried fastmode. I got the result using small samples (like with "poplistname" option), however, using all samples it does not work.

Is it impossible to?

You have enough memory for the job; maybe you've hit an EIGENSOFT fastmode bug. Try plink 2.0 "--pca approx" then.

Thank you great advice!
I'll try plink2!