lesson home

Machine Learning Methods For Material Sciences

Virtual Environments

next episode

Overview

Questions

  • What is a virtual environment?

  • How do virtual environments help manage dependencies?

  • How do you create and activate a virtual environment?


Objectives

  • Learn about virtual environments and their purpose in Python projects.

  • Understand how to create and manage virtual environments.

  • Get hands-on experience with setting up and using a virtual environment.

Python Environment Management

In ML for materials science, dependency isolation prevents conflicts between different projects. Conda and Python’s built-in venv are both used to create environments with their own Python version and libraries. Conda is preferred here due to its ability to handle scientific and compiled packages.

Conda Environments

A Conda environment is a self-contained directory with a specific Python version and packages. Do not install packages into the base Conda environment — create one per project.

Creating a Conda Environment for Materials Science ML (Linux/MacOS)

conda create --name MatML python=3.12
conda activate MatML

Installing Core ML Libraries for Materials Science

conda install numpy pandas scikit-learn matplotlib seaborn jupyterlab
pip3 install torch torchvision pymatgen ase matminer  fairchem-core fairchem-data-oc fairchem-applications-cattsunami x3dase  m3gnet matgl chgnet mp-api

Exporting and Sharing Environments

conda env export --no-builds --file MatML.yaml

Recreate the environment elsewhere:

conda env create --file MatML.yaml

Deactivating and Removing Environments

conda deactivate
conda env remove --name MatML

Virtual Environments with venv

venv is built into Python and is lighter weight but requires manual installation of all dependencies with pip.

Linux/MacOS

python3 -m venv MatML
source MatGNN/bin/activate
pip3 install torch torchvision pymatgen ase matminer  fairchem-core fairchem-data-oc fairchem-applications-cattsunami x3dase  m3gnet matgl chgnet mp-api describe numpy pandas scikit-learn matplotlib seaborn jupyterlab ipython

Windows PowerShell

python -m venv MatML
MatGNN\Scripts\activate
pip3 install torch torchvision pymatgen ase matminer  fairchem-core fairchem-data-oc fairchem-applications-cattsunami x3dase  m3gnet matgl chgnet mp-api describe numpy pandas scikit-learn matplotlib seaborn jupyterlab ipython

Deactivate with:

deactivate

Exercise: Create a MatGNN Conda Environment with ML Libraries

  1. Create and activate environment
conda create --name MatML python=3.12
conda activate MatML
  1. Install ML libraries for materials science
conda install numpy pandas scikit-learn matplotlib seaborn jupyterlab
pip3 install torch torchvision pymatgen ase matminer  fairchem-core fairchem-data-oc fairchem-applications-cattsunami x3dase  m3gnet matgl chgnet mp-api describe
  1. Verify installation
import pymatgen
import matminer
import fairchem.core
import m3gnet
import chgnet
import mp_api
import describe
import ase
print("Libraries loaded successfully")
  1. Export to YAML
conda env export --no-builds --file MatML.yaml
  1. Remove and recreate environment
conda deactivate
conda env remove --name MatML
conda env create --file MatML.yaml

Ready-to-use MatML.yaml

name: MatML
channels:
  - conda-forge
  - defaults
dependencies:
  - python=3.12
  - numpy
  - pandas
  - scikit-learn
  - matplotlib
  - seaborn
  - jupyterlab
  - pip
  - pip3:
      - torch==2.3.1
      - torchvision
      - pymatgen
      - ase
      - matminer
      - fairchem-core
      - fairchem-data-oc
      - fairchem-applications-cattsunami
      - x3dase
      - dgl -f https://data.dgl.ai/wheels/torch-2.3/repo.html
      - m3gnet
      - matgl
      - chgnet
      - mp-api
      - describe

pip3 install torch pymatgen matminer fairchem-core m3gnet matglchgnet mp-api describe numpy pandas scikit-learn matplotlib seaborn jupyterlab ipython

You can create this environment directly with:

conda env create --file MatML.yaml
conda activate MatML

Key Points

lesson home

next episode