Benchmark including ranger
and xgboost
.
batchtools.conf.R
batchtools config for running jobs locally (Multicore
orSSH
).batchmark.R
contains learner definitions with parameter spaces andbatchtools
setup viamlr3batchmark
.
The xgboost
learner seemed to consume more CPU resources than it should have been allowed to.
E.g. on a workstation with 48 threads available and and allocating 10 threads with batchtools, the load average would rise to around 48 after a few seconds.
It turns out that adding the following to .Rprofile
helped:
Sys.setenv(OMP_NUM_THREADS="1")
Sys.setenv(OPENBLAS_NUM_THREADS="1")
Sys.setenv(MKL_NUM_THREADS="1")