Codecov Report

Merging #493 into master will increase coverage by 0.94%.
The diff coverage is 92.78%.

@@            Coverage Diff             @@
##           master     #493      +/-   ##
==========================================
+ Coverage   62.02%   62.96%   +0.94%     
==========================================
  Files          59       59              
  Lines        6199     6265      +66     
  Branches      809      823      +14     
==========================================
+ Hits         3845     3945     +100     
+ Misses       2218     2181      -37     
- Partials      136      139       +3

Two things - now the font size is exceedingly big. Font size should be smaller.
Also, font type should be the same as in other Text / Orange widgets. I think this is TNR, but we use... dunno, Lato?

Overlap still occurs. Use election-tweets-2016, Preprocess Text and WC. Perhaps this can't be solved.

Also, is there any use to 'Regenerate word cloud' button? Does the visualization change at all?

I really appreciate this PR, because WC is mostly a mess. One thing that would be amazing if solved:

The 'bleeding' / protruding parts of the underlying visualization. See bottom and right side of the visualization. This is beyond annoying.

Comments:

• When I connect Topic Modeling to WC, I get green words for all topic models, even for LDA and HDP, which always give positive words. I don't mind the words being green, but if I deselect 'Color words', words should become gray.
• Sometimes I resize the window to have a nice ellipsoid word cloud. Then I select 'Regenerate word cloud' or even deselect 'Color words' (which prompts replotting) and I get the square version of WC. Very strange.
• Trying to fit all the words for, say, book-excerpts, I need 1/3 of the screen. I think this is a bit excessive. I know we talked about this, but have a look at how this appears on my screen. I have put a couple of widgets alongside for comparison. Perhaps there's no optimal solution here.

Thanks for comments:

• Topics are my fault. I forgot that the second two topics have not negative values. So now we have green/red only for the first method and normal colors for others. Also when the first method is used there is special black/white coloring for topics.
• I also realize that behavior when resizing. Resizing is handled by wordcloud2.js itself. It seems that it optimizes resizing and do not recompute all parameters on every resize. It didn't seem too problematic for me since the word cloud is staying nice.
• I decreased the size of fonts so that the cloud fit in the smaller screen area. I also added the parameter that counts word length combined with their sizes (only word length does not solve the problem when for example many words have high weights). This parameter is now included in word size calculation. The difference is the best visible when we show the topic LDA which gives similar/same values to all words (before all of them were shown with huge fonts). It works on all the examples I have tried. I hope it works on others too.