[FIX] Continuize: Disable normalizing sparse data

[VesnaT]

Issue

Fixes #4378

Description of changes

Disable Normalize by span and Normalize by standard deviation radio buttons for sparse datasets.

Includes

• Code changes
• Tests
• Documentation

[codecov]

Codecov Report

Merging #4379 into master will increase coverage by <.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #4379      +/-   ##
==========================================
+ Coverage   87.13%   87.14%   +<.01%     
==========================================
  Files         399      399              
  Lines       72901    72936      +35     
==========================================
+ Hits        63521    63557      +36     
+ Misses       9380     9379       -1

[markotoplak]

What kind of normalization can we still have with sparse data then? We could still do normalization that does some division or multiplication, we just have to avoid shifts (plus, minus).

Another thing: I never associated sparse data and discrete values, but yes, why not... Where do we get discrete sparse data in Orange? Is ti directly read from a file or generated by some text-mining widget?

[ajdapretnar]

🤔 I think that if you have some discrete variables in the corpus before bag of words, they would inevitably get transformed into sparse alongside words.

[VesnaT]

The example is described in the issue and yes, it does not make sense but it still should not crash.

[markotoplak]

Yes, but using appropriate operations for sparse data would be better than disabling options. Normalization by span is something that could still be done for sparse data, it should just not be centered.

[VesnaT]

Some types of normalization can be done using a Preprocess widget. Should these be added to the Continuize widget as well?