EXARL on Perlmutter

Load modules

module load python

and clone the repository

git clone --recursive https://github.com/exalearn/EXARL.git

Create a conda environment

conda create --name exarl python=3.9
source activate exarl

Install dependencies

conda install -c anaconda tensorflow-gpu -y
conda install -c conda-forge mpi4py gym ase lmfit scikit-learn pandas numba pybind11 pytest flake8 -y
pip install plotille

Add exarl to python path

export PYTHONPATH=<EXARL top level dir>:$PYTHONPATH

Here's an example job script for running EXARL

#!/bin/bash

#SBATCH -C gpu
#SBATCH -n 1
#SBATCH -A m3897_g
#SBATCH --gpus-per-task=1
#SBATCH -c 10
#SBATCH -t 360
#SBATCH -J ExaRL-1g-10
#SBATCH -o ExaRL-1g-10.%j.out
#SBATCH -L cfs
#SBATCH --gpu-bind=map_gpu:0

set -xe

export SLURM_CPU_BIND="cores"
export PYTHONPATH="${PYTHONPATH}:${SCRATCH}/EXARL"

srun python3 exarl/driver/ --env ExaBoosterDiscrete-v0 --n_episodes 100 --n_steps 100 --learner_procs 1
output_dir="${PWD}/exarl-output-${SLURM_JOB_ID}"