Stringent Regression Tests

[cortner]

@davkovacs

I think it would be great if we can produce a number of stringent regression tests (not part of the standard test set) where we download some existing training sets and ensure that future versions of the code can reproduce the accuracy and performance. This is not a short-term thing, but if as you say more people are starting to use ACE then we need to make sure that we really maintain the performance of the code across versions.

My group will help of course, but I think you might be the best person to produce the initial tests?

[cortner]

@gabor1 are there public links to some of your favourite training sets?

Generally - it would great to get your input on the design of these tests.

[gabor1]

we are slowly collecting some on https://libatoms.github.io/GAP/data.html

…

-- Gábor Gábor Csányi Professor of Molecular Modelling Engineering Laboratory, University of Cambridge Pembroke College Cambridge Pembroke College supports CARA. A Lifeline to Academics at Risk. http://www.cara.ngo/

On 2 Sep 2021, at 17:40, Christoph Ortner ***@***.***> wrote: @gabor1 are there public links to some of your favourite training sets? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android.

[davkovacs]

I will create some when trying to reproduce ACE1 results with ACE2.

[cortner]

@david - @andresrossb and I have started implementing performance regression tests in ACE.jl; there are some very nice packages that allow us to do this conveniently, it is worth a look. I highly recommend including this here as well. (for the performance anyhow. Haven't looked at how to do the same for model accuracy.)