Regarding StructuralHeterogeneity (structure featurizer function) of Matminer

Hello Hemanth,

This error is a result of the difficulty in producing Voronoi diagrams for crystals, and is likely caused by 1 or two different errors:

  1. Structure has atoms which are particularly far from each other. Our pymatgen implementation of the Voronoi tessellation relies on getting the atoms within a certain radius of the atom
    for which you are computing the cell. If you increase that distance, some structures will tessellate without errors.
  2. Other. There are some structures that the tessellation fails for unknown reasons. Tessellations of crystals are difficult to compute due to some numerical issues resulting from calculating
    whether >3 planes meet at a certain point, which are common in high-symmetry crystals. I’ve tried using multiple supercells of a structure or perturbing the atoms, but haven’t found anything that works consistently.

If you’d like, I can add in a setting to matminer that will allow you to increase the radius used in Voronoi calculation for StructuralHeterogeniety. In practice, I find that a few structures failing to tessellate doesn’t seriously impact
model performance.