Fig. 3: Structure-based OOD performance on Materials Project.
From: Probing out-of-distribution generalization in machine learning for materials
a Number of leave one space group out tasks with R2 higher than a given threshold as a function of threshold. The horizontal dashed line indicates the total number of tasks. LLM-Prop is not examined due to the large number of tasks and high training cost. b Leave one crystal system out performance in the ascending order of symmetry from left to right. c Leave one prototype-like out performance. Structures with fingerprints similar to a prototype are grouped together. d Leave one point group out performance in the ascending order of symmetry from left to right.
