search

MilesZhao / PGCGM

Source code for generating materials with 20 space groups using PGCGM
MIT License
28 stars 6 forks source link

Do you have a comparison to CDVAE? #2

Closed sgbaird closed 1 year ago

sgbaird commented 2 years ago

I noticed you compared results to FTCP. Do you have comparison results for CDVAE? I understand this might be less feasible for the 500k generation based on the longer time it takes to generate CDVAE samples.

usccolumbia commented 1 year ago

" I understand this might be less feasible for the 500k generation based on the longer time it takes to generate CDVAE samples". Yes you said it.

Yes. We just finished the comparison on a smaller scale. The major finding is that CDVAE is not good at generating high-symmetry crystal structures. We generated 1100 structures using CDVAE. 860 are pymatgen readable structures. Out of the top 10 space groups with most samples, 97.8% 773/790 sampleshave very low space group ID numbers (<=25). This is a serious issue as most stable structures tend to have high symmetry.

sgbaird commented 1 year ago

@usccolumbia excellent, thanks! That's great info. Generation of low-symmetry structures without explicitly enforcing it seems to be a bane https://github.com/sparks-baird/xtal2png/issues/79.