[FIX] Add labels to combo when data comes from distance matrix

[janezd]

Fixes #2863.

Includes

• Code changes
• Tests

[codecov-io]

Codecov Report

Merging #2866 into master will increase coverage by 0.02%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #2866      +/-   ##
==========================================
+ Coverage   82.06%   82.09%   +0.02%     
==========================================
  Files         327      327              
  Lines       56111    56166      +55     
==========================================
+ Hits        46046    46107      +61     
+ Misses      10065    10059       -6

[lanzagar]

In addition to fixing labels for distance matrices, this PR changes some default behaviour in the same commit.
And I am not so sure we even want to auto-set labels to the first string var. The labels a user would want are just as often in a categorical var. But even if the first string var does contain instance labels, we might not want to show them unless explcitly selected, because of our notorious label-overlapping problems. It will not look nice for any dataset with a couple hundred instances (e.g. brown-selected has short labels, <200 instances, shows the correct thing, and I find it borderline ok)

[janezd]

I'd guess that showing string labels (as opposed to showing nothing) is going to minimize the expected number of mouse clicks. If I have string labels - and I'll often have a single one, I'd usually want to see them.

For class labels, we already have colors. For attribute labels, we can't guess which one to show and we'll usually miss.

@BlazZupan?

But I agree that we shouldn't show labels by default if there are too many. (You already said this on Friday, but I forgot.) Should I add an and len(data) <= 200 above?

[lanzagar]

A popular thing from text mining, which we showed several times in workshops (!), is to load a corpus (use the default book-excerpts), connect to Bag of Words and then to MDS to show a map of documents.
Try that after this PR ;)

I am still in favour of not showing any labels automatically, but if I had to, then in addition to len(data)<=200 I would add that the max len of all string values from the variable should be <12 or something like that.

[janezd]

I removed default labels. I haven't squashed the commits, so we can bring them back (now or in the future).