Tomoki-YAMASHITA / CrySPY

CrySPY is a crystal structure prediction tool written in Python.
https://tomoki-yamashita.github.io/CrySPY_doc
MIT License
107 stars 38 forks source link

How to calculate RMSE for a predicted a structure and a true structure? #5

Closed usccolumbia closed 3 years ago

usccolumbia commented 4 years ago

Usually, we need to do some kind of evaluation to check the performance of the CSP algorithm. This can be done by calculating the RMSE distance between the predicted structure and the predicted structure... Does CrySPY has such a function?

Tomoki-YAMASHITA commented 4 years ago

No, but CrySPY will be able to output predictive values and variance in BO. I think we can also include RMSE.

usccolumbia commented 4 years ago

If you can add the RMSE function to it, it will be great to evalute the performance of diff algorithms... a reference is here https://pyxtal.readthedocs.io/en/latest/_modules/pyxtal/operations.html#distance_matrix

Tomoki-YAMASHITA commented 4 years ago

Even if we evaluate the structure distance between predictive and true structures (or evaluate energy difference using RMSE), we would not be able to evaluate the performance of CSP algorithms. That is only the evaluation of accuracy of the machine learning potential. CSP includes structure generation and exploration phase.

sgbaird commented 2 years ago

Related: https://github.com/sparks-baird/xtal2png/issues/12

See also pymatgen's StructureMatcher and get_rms_dist function.