librascal
librascal copied to clipboard
Rename SparsePoint to something else
We are using the historical name of sparse points when working with GAP potentials. However, this name can be confusing since the "sparse" kernel is not a sparse matrix in the usual sense, only a low-rank one.
I think we should change the name to something clearer. One alternative is Pseudo-points, which have been used a lot in the group already, but @max-veit does not like it (I'll leave him explain why below!).
Another alternative is basis points, which is very clear in the GPR framework, although I thing basis is already used a lot in this code.
I personally like support point quite a lot: it tells what it is without being too ambiguous. The only source of confusion I could see is with support vector machines.
Do you have better alternatives, or arguments for keeping the "sparse point" naming?
You can add 'inducing' points to the list of potential rename. That being said I would prefer to avoid inventing a new name since the littérature already accepts a few names for the same object (support and basis have not been used as far as I know). Moreover I would not rename this object either since this is one of the accepted terminology so it's clear enough what this is.
The terminology of 'active points' is also gaining currency FWIW. But personally I don't see a problem with calling them sparse points. The linear-algebra sense of 'sparsity' is a very narrow, technical definition that doesn't have the same meaning in other fields; it makes perfect sense to call the GPR active/inducing points 'sparse' since they're usually chosen (sparsely!) from the training data.