[REVIEW] Keep single-series DataFrames as DataFrames in preprocessing

[wphicks]

Ensure that single-series DataFrames are not converted to Series during preprocessing

Resolve #3059

[GPUtester]

Please update the changelog in order to start CI tests.

View the gpuCI docs here.

[wphicks]

@viclafargue Would you mind taking a look at this one?

[viclafargue]

Thank you very much for tracking down the bug in #3059! I'm just a bit concerned about the possible case in which one preprocessor can receive a Series but has to output a Dataframe. But if all tests pass, it should be fine. Otherwise, LGTM !

[wphicks]

Thanks for the review @viclafargue! To make sure I understand, you're thinking about a preprocessor that takes in a single Series but then blows it out into something 2D, right? That is indeed something I hadn't thought about. Let me update to_output_type as well to handle that case appropriately; otherwise we'd get a ValueError from the CumlArray code.

[wphicks]

@viclafargue Mind taking a look at the update? I think this should handle the case you pointed out without requiring further specialized code in each preprocessor.

[viclafargue]

@viclafargue Mind taking a look at the update? I think this should handle the case you pointed out without requiring further specialized code in each preprocessor.

The update and its testing look good to me. Thanks @wphicks!

[wphicks]

rerun tests

[codecov-io]

Codecov Report

Merging #3069 (b3c2719) into branch-0.17 (25b29f5) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@             Coverage Diff              @@
##           branch-0.17    #3069   +/-   ##
============================================
  Coverage        69.38%   69.39%           
============================================
  Files              193      193           
  Lines            14704    14708    +4     
============================================
+ Hits             10203    10207    +4     
  Misses            4501     4501           

Impacted Files	Coverage Δ
...on/cuml/_thirdparty/sklearn/preprocessing/_data.py	63.54% <100.00%> (+0.08%)	⬆️
python/cuml/thirdparty_adapters/adapters.py	88.55% <100.00%> (+0.11%)	⬆️

---

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 25b29f5...b3c2719. Read the comment docs.

[wphicks]

Looks like that test failure may be genuine. I'll look into it and update.

[wphicks]

I was not able to reproduce this locally with my existing conda env, so I'm suspicious of an upstream change. I've been trying to update my local conda env, but it keeps hanging indefinitely. I'll have to take a closer look a little bit later.

[wphicks]

Finally fixed the error that CI picked up. I tweaked the normalize function to prevent returned norms from being coerced to DataFrames unnecessarily.

@viclafargue Would you mind doing a re-review on this one?

[viclafargue]

It looks good to me! Thanks for working on that @wphicks