[FIX] owpredicttions: Remove special magic value 2

[ales-erjavec]

Issue

Fixes gh-6911

@janezd what was the point of using 2 as invalid value. I cannot find any place where it would be used/checked,
except where it is replaced with NaN on output.

Description of changes

Remove special magic value 2. Just use NaN.

Includes

• Code changes
• Tests
• Documentation

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 88.46%. Comparing base (871b406) to head (de89d5f).
Report is 52 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #6933      +/-   ##
==========================================
+ Coverage   88.39%   88.46%   +0.06%     
==========================================
  Files         329      331       +2     
  Lines       72480    73097     +617     
==========================================
+ Hits        64069    64663     +594     
- Misses       8411     8434      +23     

---- 🚨 Try these New Features:

• Flaky Tests Detection - Detect and resolve failed and flaky tests

[janezd]

@janezd what was the point of using 2 as invalid value.

I'm guessing: the error column contained 2 as the error in probability estimation for discrete problems. In discrete data, this was used only for sorting. I probably wanted to put those values last, hence 2 made sense.

This became a problem after 2375dc7, which started using errorColumn for output.

This fix is OK if we can assume that numpy treats nan's as larger than ordinary numbers and will continue to do so, although I can't find this in documentation.

An alternative would be to do err[err == 2] = numpy.nan only for classification.

But the assumption is probably going to hold and your fix makes the code simpler.