ml-mlip-benchmark

What problem does it solve?

This Skill measures how accurately a machine learning interatomic potential (MLIP) reproduces a labeled ground-truth dataset by computing standard error metrics and producing parity plots for rapid visual validation.

Core Features & Use Cases

• Benchmark accuracy with MAE/RMSE: Computes Mean Absolute Error and Root Mean Square Error for energy-per-atom and atomic forces, with optional stress support.
• Generate parity plots: Produces energy and forces parity plots (and stress parity when available) to quickly spot systematic bias or outliers.
• Use with labeled dataset formats: Works with a JSON dataset whose entries include a structural dictionary plus ground-truth fields for energy/forces (and optionally stress) matching the format used by related MLIP training data.

Quick Start

Run the benchmark with the labeled dataset JSON you have and save the resulting metrics and prediction parity data to a results JSON file.