Closed MatthiasSachs closed 3 years ago
thank you again! I'll review asap. Liwei is also almost ready, so it this is now getting exciting!
can we remove the dependency on ReferenceFrameRotations
and just use randomly generated rotation matrices?
Sure!
looks like basis generation is very slow, correct? we should hunt for some type instabilities
can you please give me push access to your repository then I can push some small changes. I just adpated your tests to the most recent changes in ACE.jl
but great thing is that all tests appear to pass ok!!
I just added you as a collaborator to my forked version of the repository
I guess this is almost ready to merge. Just want to spend a bit more time reading through your changes. Thank you!
With regards to the speed issue: I have not been focusing on speed yet. I am wondering if we could maybe more effectively make use of the orthogonality of the radial basis functions? If I recall correctly we are executing the "symmetrization operations" multiple times when we form the basis?
Correct but that’s also the case for invariants. This seems worse. But you are right we can push the performance back a a bit
I've made various comments. Sorry for cluttering this conversation, I should have created a single "review". Anyhow, would you be willing to take care of these points and push final changes? Then I'd probably like to have a conversation about merging also with Liwei's branch and in which order.
ok, I am merging this now, but see new items in #23
New pull request:
First working version for EuclidianVector-basis
Relevant tests are provided in test/test_symmbasis-equiv.jl
The test sets is not yet added to the general test set.