Probabalistic metric space

[redhog]

Point pairs can be sampled randomly. Two random subset of all points are chosen, and the distance matrix is calculated only between these two subsets. The size of each subset is set by samples: if < 1 it specifies a fraction of all points, if >= 1 it specifies the number of points in each subset.

Closes #48

Only look at this after merging #84

[mmaelicke]

wow it's going really fast here...
I'll try to keep up with your pace :)

[redhog]

This is all stuff I'm building for a single feature in our software. For a bit of background: We have built (and are using) https://github.com/emerald-geomodelling/EmeraldTriangles to manage triangulated datasets. Among those are topography, and while that usually is delivered as a raster (geotiff) and easy to sample to the TIN (which is typically build from a grid + all measurement points), sometimes it comes as an existing TIN that we just add the new (grid + measurement) points to. In that case I wanted to use kriging on topo too, not just measurements. But for such a topo (DTM) TIN, the number of points is very large, and I ran into all the things I've been addressing in these two PRs.

[redhog]

For now the scikit-gstat kriging is done outside of EmeraldTriangles, but my goal is to have a utility function in that library that uses vertice points with a non-nan value in some column as observations, kriging to all rows with nan values in the same column.

[codecov]

Codecov Report

Merging #87 (7b63fb9) into master (6d972e9) will increase coverage by 0.08%.
The diff coverage is 93.84%.

@@            Coverage Diff             @@
##           master      #87      +/-   ##
==========================================
+ Coverage   90.92%   91.01%   +0.08%     
==========================================
  Files          15       15              
  Lines        1785     1847      +62     
==========================================
+ Hits         1623     1681      +58     
- Misses        162      166       +4     

Impacted Files	Coverage Δ
skgstat/MetricSpace.py	87.75% <93.33%> (+3.84%)	⬆️
skgstat/Variogram.py	97.05% <100.00%> (+0.01%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 6d972e9...7b63fb9. Read the comment docs.

[redhog]

@mmaelicke This is ready for merging now whenever you have time :)

[mmaelicke]

@redhog, looks good to me, nice extension.
I think with two unittests we can bring the coverage back. Both could use the Variogram to instantiate a ProbabilisticMetricSpace with two different sample sizes. That should cover all the new code.
If you want me to do that, just go ahead and assign me to the PR, I think I will find some time towards the end of the week.

[redhog]

Btw, @mmaelicke I saw you added me to the citation! Thanks! I made an account with orcid so maybe you coul update the author list on zenodo there to point to my id https://orcid.org/0000-0002-8254-1163?

[mmaelicke]

@redhog, thanks for adding all the stuff, I think everything is addressed. There is one more thing that came to my mind, should be an easy fix
Best

[mmaelicke]

Thanks! Great PR, I really like it.

[mmaelicke]

You can merge whenever you want. I will add the changelog and release the new version to PyPI, this afternoon