search

Vivianstats / scImpute

Accurate and robust imputation of scRNA-seq data
https://www.nature.com/articles/s41467-018-03405-7
90 stars 34 forks source link

doParallel performance on windows edition R #24

Open ygidtu opened 5 years ago

ygidtu commented 5 years ago

When I try to use scImpute, the ncores is useless. It will use all the CPU resources automatically.

As far as I know, Windows system has different parallel mechanism with Unix-based system like: Linux and OSX. Therefore the doParallel package design different usages for Windows and Unix. (Please see: https://stackoverflow.com/questions/45819337/option-cores-from-package-doparallel-useless-on-windows).

And the point is the R installed by conda is a Windows edition. So I compared the difference between system default R (3.5.2; Ubuntu 16.04) and conda installed R (3.5.1; Platform x86_64-conda_cos6-linux-gnu).

image

So can you check how to avoid this issue?

aziway53 commented 5 years ago

git1 "ncores = 1" is working on windows, is it ok?

ygidtu commented 5 years ago

git1 "ncores = 1" is working on windows, is it ok?

I didn't test on a Windows version of R, I just very positive this is due to conda R. And the conda R is not regular version of R, it's microsoft enhanced distribution of R (MRAN: https://mran.microsoft.com/open). So I guess this is the reason