Extension of spatial.distance module

[Cdebus]

• refactorization to scipy naming
• restructuring of distance calculation logic
• support for distance calculation of two different tensors with different splits
• support for dtypes other than float32

Description

Issue/s resolved:
Addresses #475 amongst other changes

Changes proposed:

• Implementation of abstract function _dist(X,Y, metric)
• derive cdist and rbf as explicit implementatiions with metrices euclidian and gaussian

Type of change

• New feature (non-breaking change which adds functionality)

Due Diligence

• All split configurations tested
• Multiple dtypes tested in relavent functions
• Documentation updated (if needed)
• Updated changelog.md under the title "Pending Additions"

Does this change modify the behaviour of other functions? If so, which?

no

[codecov]

Codecov Report

Merging #479 into master will increase coverage by 0.03%.
The diff coverage is 98.48%.

@@            Coverage Diff             @@
##           master     #479      +/-   ##
==========================================
+ Coverage   96.72%   96.76%   +0.03%     
==========================================
  Files          60       60              
  Lines       12281    12562     +281     
==========================================
+ Hits        11879    12155     +276     
- Misses        402      407       +5

Impacted Files	Coverage Δ
heat/cluster/kmeans.py	82.22% <100%> (ø)	⬆️
heat/spatial/__init__.py	100% <100%> (ø)	⬆️
heat/spatial/tests/test_distances.py	96.96% <100%> (+6.77%)	⬆️
heat/spatial/distance.py	97.34% <97.34%> (ø)

---

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 29be6aa...a65b5cc. Read the comment docs.

[coquelin77]

looks pretty good. needs a few small updates but good job!

[coquelin77]

device argument

[coquelin77]

would it be possible to abstract the communications loop that you use in _dist for the circular sending/receiving?

[Cdebus]

would it be possible to abstract the communications loop that you use in _dist for the circular sending/receiving?

I don't know, in principle yes, but what would be the application. The key point here ist the calculation of metrics between chunks from X and Y. While this is easily substituted, it can already currently be modified by the function you pass as metric.