Optimize distance function in kriging

[redhog]

When krigin, a distance function is called for each point to calculate its distance to every point in the variogrammed dataset. This might be necessary if the use has supplied their own distance metric, but if it's one of the built in scipy, we can use scipy.spatial.distance.cdist on all points instead, and if it's euclidean, we can even use scipy.spatial.cKDTree, see ttps://iliauk.wordpress.com/2016/02/16/millions-of-distances-high-performance-python/ , which would automatically cull the points farther away than the range without actually doing a distance calculation on them.

• Optimize observation selection
• Optimize kriging equation creation

[mmaelicke]

I commented in #60 ...

[mmaelicke]

@redhog was this resolved by #60 ?

[redhog]

Only half-way. I added the two subtasks above...

But the solution is to use the same method as with #60 above, including the cKDTree optimization.

[mmaelicke]

OK, great! Then I'll leave this open.

[redhog]

The idea I have is to generate a pdist matrix for all observation points at the top, then select the sub-matrix for the nearby points instead of calling self.dist().
This might even be worth allowing "short circuit" for between Kriging and Variogram, that is, allow you to send in a distance matrix alongside with coordinates and values.

[redhog]

As far as I can see this is solved in https://github.com/mmaelicke/scikit-gstat/blob/master/skgstat/Kriging.py#L415