Skip to content

Machine Learning#

Machine learning refers to a set of techniques and algorithms that enable computers to automatically learn from data and improve their performance on a specific task over time. Types of machine learning methods include, but are not limited to, supervised learning (algorithms trained on labeled datasets), unsupervised learning (algorithms trained on unlabeled datasets), and reinforcement learning (learning by trial and error). The Computational Science Center at NREL conducts research in these types of machine learning, and also supports the use of machine learning software on Kestrel.

Getting Started#

This section provides basic examples for getting started with two popular machine learning libraries: PyTorch and TensorFlow. Both examples use Anaconda environments, so if you are not familiar with their use please refer to the NREL HPC page on using Conda environments and also the Conda guide to managing environments.

Getting started with PyTorch#

To begin, we will outline basic steps for building a simple CPU-based conda environment for PyTorch. First, load the anaconda module and create a new conda environment:

module load anaconda3

Answer yes to proceed, and you should end up with directions for starting your conda environment pt. Note that these instructions place your environment in the specified /projects folder. This is advisable, as opposed to installing conda environments in their default location in your home directory. See our Conda documentation for more information.

Activate the pt conda environment and install PyTorch into the active conda environment:


conda install pytorch torchvision torchaudio cpuonly -c pytorch
Answer yes to proceed, and you should be up and running with PyTorch! The PyTorch webpage has great resources for getting started, including resources on learning the basics and PyTorch recipes.

Getting started with TensorFlow#

Getting started with TensorFlow is similar to the process for PyTorch. The first step is to construct an empty conda environment to work in:

module load anaconda3

Subsequently, activate the tf conda environment, ensure you are running the latest version of pip in your environment, and install the CPU only version of TensorFlow using pip:
pip install --upgrade pip
pip install tensorflow-cpu
You should now be up and running with a TensorFlow! Similar to PyTorch, the TensorFlow webpage has lots of great resources for getting started, including turotials, basic examples, and more!

Example Job Script#

PyTorch or TensorFlow CPU
#SBATCH --nodes=1           # Run the tasks on the same node
#SBATCH --time=1:00:00          # Required, estimate 1 hour
#SBATCH --account=<your_account>
#SBATCH --exclusive         # if you want to use the whole node

module load anaconda3 

cd /projects/<your_project_here>/<your_code_directory>

conda activate /projects/<your_project_here>/<folder_for_conda_envs>/pt #or tf

srun python


This Getting Started section is only scratching the surface of ML libraries and resources that can be used on Kestrel. Tools such as LightGBM, XGBoost, and scikit-learn work well with conda environments, and other tools such as Flux for the Julia Language can be used on Kestrel as well.

Once you have completed your batch file, submit using

sbatch <your_batch_file_name>.sb


The above examples are simple CPU-based computing environments. To build conda environments for GPUs we refer you to the PyTorch and TensorFlow install directions.

For optimized TensorFlow performance, we recommend using a containerized version of TensorFlow.