[MRG] Fix SmoteNC zero variance resampling

[freddyaboulton]

Reference Issue

Fixes #837

What does this implement/fix? Explain your changes.

Fixes the issue as described by @glemaitre here: #837 (comment)

Any other comments?

Had to add ytype to base class generate_samples api in order to know which class we're resampling so we can use the right subset of _X_categorical_minority_encoded

[codecov]

Codecov Report

Merging #905 (ca2935a) into master (6176807) will increase coverage by 0.03%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #905      +/-   ##
==========================================
+ Coverage   97.19%   97.23%   +0.03%     
==========================================
  Files          97       97              
  Lines        6348     6361      +13     
  Branches      711      712       +1     
==========================================
+ Hits         6170     6185      +15     
+ Misses        106      105       -1     
+ Partials       72       71       -1     

Impacted Files	Coverage Δ
imblearn/over_sampling/_smote/base.py	99.02% <100.00%> (+0.01%)	⬆️
imblearn/over_sampling/tests/test_common.py	100.00% <100.00%> (ø)
imblearn/keras/_generator.py	94.73% <0.00%> (+2.63%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 6176807...ca2935a. Read the comment docs.

[glemaitre]

Thanks. Since I remade the implementation of SMOTE to compute the median on each subclass, I added your bug fix (or something similar). Thanks @freddyaboulton

Closing this PR.

[freddyaboulton]

Thanks @glemaitre !