cluster based on fingerprint distance and composition type

jlaw9 / crystal-gnn

Work in progress on graph neural networks for crystal energy prediction

0 stars 0 forks source link

Closed pstjohn closed 1 year ago

pstjohn commented 2 years ago

[ ] we should restrict the clustering by composition type
[ ] cluster those that don't match known prototypes to find potential new prototypes.
[ ] if multiple calculations converge to the same prototype, keep the one with the lower energy?