Skip to content

Latest commit

 

History

History
44 lines (19 loc) · 985 Bytes

README.md

File metadata and controls

44 lines (19 loc) · 985 Bytes

Warning opnmfR is still under development and some parts are not completely tested.

Installation

In R terminal

install.packages("remotes")
remotes::install_github("kaurao/opnmfR")

Note Microsoft R Open comes with MKL and can substantially improve runtime performance of Rcpp functions by using threads.

Use it

data("iris")

nn = opnmfR_ranksel_perm(data.matrix(iris[,1:4]), 1:4, W0="nndsvd")

Using gpuR (non tested)

Install gpuR following the instrctions here:

Note that some of the instructions are opudated, e.g. nvidia toolkit for ubuntu14.04. Go to the nvidia page and follow the instructions there.

I needed to do the following on ubuntu 16.04:

sudo apt-get install opencl-headers ocl-icd-opencl-dev

Get the correct .deb from https://developer.nvidia.com/cuda-downloads and follow the associated instructions.

Run sudo R and then install.packages('gpuR')

And then run R in sudo mode