Examples Overview

In this section, we demonstrate the capabilities and usage of the Orchestrator in multiple contexts, both as standalone modules and in combination for more complex workflows. To make these examples accessible, they are provided as Jupyter notebooks with in-line comments that explain conventions and typical behaviors. In addition to these examples, you can refer to the unit test functions found in orchestrator/test/<MODULE_NAME>/<MODULE_NAME>_unit_testers.py files for further insights into module usage. The range of these tests are enumerated in the Testing section of the docs.

Jupyter Examples

The following notebooks are designed to be self-contained and can be explored in any order. Depending on the example, you may need to perform some initial setup, such as adding configurations to the storage module or installing a KIM potential from openkim.org.

Pruning a Dataset with the Augmentor

This notebook walks through the process of computing QUESTSDescriptor descriptors on a dataset, followed by pruning the dataset based on these descriptors using the Augmentor module. After pruning, you will see how to train models on both the original and reduced datasets using the FitSnapPotential and FitSnapTrainer, allowing you to compare their performance.

Download Notebook

Train a Potential and Deploy it in a MD Simulation

Here, you will learn how to train a FitSnapPotential IAP from scratch. The example continues by showing how to save and install the trained model to the KIM API, and then demonstrates how to run a molecular dynamics simulation with the new potential using a LAMMPSSimulator.

Download Notebook

Use a Potential to Evaluate a Material Property

In this example, you will see how to initialize a KIMPotential and use it to evaluate material properties. The notebook guides you through generating a cold curve using both the potential’s evaluate() method and the KIMRun target property, illustrating different approaches to property evaluation.

Download Notebook

Use Score Modules to Analyze a Dataset

This notebook demonstrates how to load a dataset—either from storage or from an xyz file—and then compute descriptors for further analysis. You will then see how to assess the dataset using various scoring modules, including QUESTSEfficiencyScore, QUESTSDiversityScore, and QUESTSDeltaEntropyScore calculations.

Download Notebook

Refit an Empirical Potential

Starting from an existing empirical potential, this example guides you through evaluating and inspecting its parameters. You will then set up a ParametricModelTrainer to retrain the model, and finally re-evaluate and save the updated potential, completing the refitting workflow.

Download Notebook