search

ACEsuit / ACEpotentials.jl

Machine Learning Interatomic Potentials with the Atomic Cluster Expansion
MIT License
45 stars 11 forks source link

Purify 2B in new kernels #228

Open cortner opened 3 weeks ago

cortner commented 3 weeks ago

@CheukHinHoJerry -- I think before we can get rid of ACE1 and ACE1x we may want to implement at least the 2B purification in the new implementation. Do you have any thoughts on that? Can we both think about it and discuss early next week?

CheukHinHoJerry commented 3 weeks ago

Sure - I will mull over this during weekend.

cortner commented 1 week ago

@CheukHinHoJerry -- I think this is the only missing feature for 0.8. Have you managed to give any thought to this yet? Don't worry if not, but I don't want to start if you have already done something.

CheukHinHoJerry commented 1 week ago

I haven't done anything on this yet. We discussed last time that we are more inclined to do this as a separate lux layer. I am still uncertain on how to do this efficiently but maybe we shouldn't worry about the cost and just do this naively?

cortner commented 1 week ago

Let me see if I find any time to have a first go at this over the next two days.

cortner commented 6 days ago

just to record here that there is a draft in P4ML, but it seems to be a little non-trivial to make this both efficient and general (i.e. non-polynomials...)