Search code examples
rdeep-learningmxnet

mxnet multicore on LInux in R

Follwoing works fine on windows although throws error on Linux

core <- 5
ctx <- lapply(c(1:core), function(i) {mx.cpu(i)})

The model function is "mx.model.FeedForward.create"

The error is "Error in (function (symbol, ctx, grad.req = "null", ...) : Need more shape information to decide the shapes of arguments"

Although, it works properly when core <- 1

Solution

  • Here is solution.

    First set the environment variable

    Sys.setenv(MXNET_CPU_WORKER_NTHREADS = core);

    Use ctx= mx.cpu() depending upon cpu/gpu

    It is working on both Windows and Linux