[ENH] Dask: Logistic Regression

[noahnovsak]

LogisticRegressionLearner and the corresponding widget now support DaskTables. Currently only for binary problems (*filtering rows doesn't work, because it doesn't update the target variable's list of possible values, to fix this we would need to check the contents the table instead).

Changes:

• LogisticRegressionLearner imports dask_ml which we may not want to add to the requirements
• _initialized_wrapped method of SklLearner now takes two arguments
• DaskTable overrides _filter_has_class method to avoid using bottleneck on dask (assumes DaskTables are never sparse)
• added np.asarray in a number of places to compute values when it seemed reasonable (statistics, predictions, test & score)

[codecov]

Codecov Report

Merging #6512 (fd1a466) into dask (89e5caf) will decrease coverage by 0.01%.
The diff coverage is 81.48%.

Additional details and impacted files

@@            Coverage Diff             @@
##             dask    #6512      +/-   ##
==========================================
- Coverage   87.70%   87.70%   -0.01%     
==========================================
  Files         322      322              
  Lines       69981    70004      +23     
==========================================
+ Hits        61377    61395      +18     
- Misses       8604     8609       +5