Use the new MinkowskiMetric type from Distances.jl

[eliascarv]

No description provided.

[dkarrasch]

In your local copy, you'll need to switch the environment to benchmark and then benchmark> up (in pkg mode) update the manifest (otherwise it will not catch the latest Distances.jl) and then push the changes to Manifest.toml to this branch.

[eliascarv]

Done!

[eliascarv]

@dkarrasch, I just noticed something while reading the KDTree code: the implementation uses functions from the UnionMetric API, which are: eval_reduce and eval_end. In other words, it will not work with any MinkowskiMetric.
That is, only UnionMinkowskiMetric are supported in the KDTree code.

[eliascarv]

@dkarrasch, what do you think about moving the eval_pow and eval_diff functions to Distances.jl?
These functions seem to be part of the UnionMetrics API.
Or do you think these functions are too specific and are only for NearestNeighbors.jl internal use?

[eliascarv]

@KristofferC, can you review this PR please? It has already been approved by Daniel.

[dkarrasch]

@dkarrasch, what do you think about moving the eval_pow and eval_diff functions to Distances.jl? These functions seem to be part of the UnionMetrics API. Or do you think these functions are too specific and are only for NearestNeighbors.jl internal use?

I'm not sure. Aren't they only used for UnionMinkowskiMetrics? The other thing is that there is no documented UnionMetrics API, right? You need to distill it from the Distances.jl code.

I think one would need concrete reasons to change something that has been up and running for so long. The risk of breakage is too big, I'd say.

[juliohm]

ping @KristofferC

[eliascarv]

@KristofferC, this PR is ready. Can you review it?

[juliohm]

Is there anything we can do to get this merged? @KristofferC could you give @dkarrasch permissions to maintain this package?

[KristofferC]

All I can do is blame the horrible github notification UI for not showing me this earlier.