⚛️🏋 matbench-genmetrics 🎨📐
Many generative models for crystal structure have been developed; however, few standardized benchmarks exist, and none exist in a format as easy-to-use as Matbench. Here, we introduce matbench-genmetrics, an open-source Python library for benchmarking generative models for crystal structures. We incorporate benchmark datasets, splits, and metrics inspired by Crystal Diffusion Variational AutoEncoder (CDVAE), which has been demonstrated as state-of-the-art in generative crystal structure tasks. We provide our own benchmarks using time-series style cross-validation splits from Materials Project via our mp-time-split package and focus on four metrics: validity, coverage, novelty, and uniqueness.
Star
Follow @sgbaird
Issue
Discuss