[ENH] OWTreeViewer: Bold predicted values in tree nodes

[irgolic]

Description of changes

I think this better visually differentiates a node's prediction from its next branching feature.

On a tangent, is the background color of a regression tree's nodes always this color by default? Either the background's or the separator line's color should be changed.

Includes

• Code changes
• Tests
• Documentation

[codecov]

Codecov Report

Merging #4269 into master will decrease coverage by <.01%.
The diff coverage is 50%.

@@            Coverage Diff             @@
##           master    #4269      +/-   ##
==========================================
- Coverage   86.24%   86.24%   -0.01%     
==========================================
  Files         396      396              
  Lines       70714    70714              
==========================================
- Hits        60989    60987       -2     
- Misses       9725     9727       +2

[ajdapretnar]

You mean is the background always purple? Yes and no. You can set the node color in the Tree Viewer. By default it is colored by, unremarkably, default color, but you can change it to mean value and such. Perhaps the default is not the best... well, default.

[irgolic]

@ajdapretnar I'll look into changing the default color, or maybe I'll make the horizontal line white.

On another note, in the regression tree, should we bold the errors if they're significant?

[ajdapretnar]

On another note, in the regression tree, should we bold the errors if they're significant?

I like the idea, but would they stand out enough considering the mean value is already bold?

[irgolic]

Also, can anyone explain why this PR is failing codecov?

[janezd]

Also, can anyone explain why this PR is failing codecov?

The second of the two lines you've changed is not covered by tests, apparently. The coverage is not worse, but the diff coverage is < 95%.

Colors are indeed ugly.

As for marking significant differences, I don't like it because (a) it doesn't stand out and thus looks more like an inconsistency, (b) the user won't know why are they bold anyway, (c) significances could only be computed on separate data set, otherwise this is pure overfitting. Especially (c) is a problem.