Request: element removal option for neighbors.KDTree

[mwharton3]

It would be useful to be able to delete elements from a previously constructed KD-tree. This would be a method of the KDTree class, i.e. KDTree.delete(index). This could save time in constructing a new KD tree in the case of, for example, decimation.

This is outlined in the original KD-tree paper (Bentley) in section 5.

[jnothman]

I'm not sure this is a common machine learning use case, so we are not likely to prioritise it being implemented

[yashjakhotiya]

If this has not been assigned yet, I would like to work on it @jnothman

[rth]

@yashjakhotiya You can certainly evaluate how much effort / code changes this feature would require.

However there this no guarantee that this feature would be accepted, particularly if it involves adding (and maintaining) as significant amount of code. Scipy might be actually a better place for it (see scipy/scipy#9029) unless there is an important use case in some popular or classical ML algorithm I'm not aware about.

[mwharton3]

I had done some tests using KD-trees a couple months ago with 3 dimensional data, I found the sklearn framework to be quite a bit faster than scipy. i.e. anyone with a large dataset that wanted this feature would find more benefit here.