[ENH] Gradient Boosting: New widget

[VesnaT]

Issue

Implements #5107
Alternative to #5116

Description of changes

• wrap sklearn gradient boosting classification and regression learners
• wrap catboost classification and regression learners
• wrap xgboost classification and regression learners
• Gradient Boosting widget

Includes

• Code changes
• Tests
• Documentation

[codecov]

Codecov Report

Merging #5160 (bfffd6e) into master (ef789ee) will increase coverage by 0.20%.
The diff coverage is 97.80%.

@@            Coverage Diff             @@
##           master    #5160      +/-   ##
==========================================
+ Coverage   84.77%   84.98%   +0.20%     
==========================================
  Files         286      296      +10     
  Lines       60084    60643     +559     
==========================================
+ Hits        50938    51539     +601     
+ Misses       9146     9104      -42     

[lanzagar]

Works well for me.
I would suggest:

• being explicit for the default "Gradient Boosting" method as well and changing to "Gradient Boosting (scikit-learn)" to be more consistent with other entries in the dropdown
• changing the default num of trees for catboost from 500 to 100 so it is the same as all the other methods - catboost is already the slowest (might need additional investigation)
• change default value of Replicable training to True. BTW is this really only available from scikit-learn, do xgboost and catboost not allow setting random seeds?

[VesnaT]

• change default value of Replicable training to True. BTW is this really only available from scikit-learn, do xgboost and catboost not allow setting random seeds?

I think so. Both (xgboost and catboost) have 0 as default. If one passes None, it is changed to 0.