[FIX] Round bhattacharayya

[AndrejaKovacic]

Issue

Due to floating point precision, bhattacharayya distance would sometimes return negative distances.

Description of changes

Rounding numbers to 13th digit. I think this is precise enough for our purposes, otherwise, we could use decimal class instead of floats.

Includes

• Code changes

[codecov]

Codecov Report

Merging #4340 into master will increase coverage by 0.06%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #4340      +/-   ##
==========================================
+ Coverage   86.85%   86.91%   +0.06%     
==========================================
  Files         396      396              
  Lines       71828    71991     +163     
==========================================
+ Hits        62383    62571     +188     
+ Misses       9445     9420      -25

[janezd]

I think this should be solved by clipping, not rounding.

[janezd]

I know I'm being an a..., I mean, annoying, but I would find it much nicer if you changed the code to something like

if sp.issparse(a):
    prod = a.multiply(b)
else:
    prod = a * b
return np.clip(-np.log(np.sum(np.sqrt(prod))

Or even prod = a.multiply(a) if sp.issparse(a) else a * b.

[AndrejaKovacic]

it does look nicer, i changed it.

[markotoplak]

Hmmm, is the condition even needed? Wouldn't a.multiply(b) work to with sparse and dense matrices?

[janezd]

You seem to think I haven't tried. :)

>>> import numpy as np
>>> a = np.arange(20).reshape(4, 5)
>>> a.multiply(a)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
AttributeError: 'numpy.ndarray' object has no attribute 'multiply'

There is np.multiply, but it doesn't work for sparse matrices.

[AndrejaKovacic]

No, numpy arrays don't have multiply function, only the numpy module has it.