Rename CV and TrainTest "stratification" parameter

[rogancarr]

This PR changes the CrossValidation and TrainTest parameter StratificationColumn to be GroupPreservationColumn and updates the docstrings to give a clearer explanation.

Fixes #2536

See conversation in #2536

Related to #1204, but that issue might be asking for further coverage.

[rogancarr]

Something has gone weird with my master branch, and I'm seeing every merge from upstream as a commit. I'll look into this, but let's work with it for now ;)

[codecov]

Codecov Report

Merging #2537 into master will increase coverage by 0.17%.
The diff coverage is 76.19%.

@@            Coverage Diff             @@
##           master    #2537      +/-   ##
==========================================
+ Coverage   71.26%   71.44%   +0.17%     
==========================================
  Files         797      801       +4     
  Lines      141292   141855     +563     
  Branches    16118    16141      +23     
==========================================
+ Hits       100692   101346     +654     
+ Misses      36138    36041      -97     
- Partials     4462     4468       +6

Flag	Coverage Δ
#Debug	71.44% <76.19%> (+0.17%)	⬆️
#production	67.73% <73.68%> (+0.14%)	⬆️
#test	85.53% <100%> (+0.17%)	⬆️

[Ivanidzo4ka]

I think you left that description from previous iteration with idColumn.
Do you want to rephrase it, since ID looks weird. #Resolved

[TomFinley]

Also ID is already a thing on IDs, naming something different that would be super confusing.

machinelearning/src/Microsoft.Data.DataView/IDataView.cs

Line 132 in 1f29d2c

public abstract ValueGetter<DataViewRowId> GetIdGetter();

[TomFinley]

I @rogancarr , thanks for looking at this.

Group column is already an established name used to describe actual things that are intended to form a group. This is more general than this. So I do not like the name "group preservation." It's far too close. "Hey I specified this group column, how come my ranking metrics are all funky?" "Ah, you did not specify the group column, but the group preservation column!" Looks silly.

I'd almost prefer a random name drawn out of a dictionary to this. That would just look weird, but this will just be consistently confusing. I'll post suggestions in issue.

[TomFinley]

Something has gone weird with my master branch, and I'm seeing every merge from upstream as a commit. I'll look into this, but let's work with it for now ;)

You commited directly into master branch on your own fork about a week ago, somehow didn't catch it, pushed that cahnge, and you've been doing pulls ever since.

Once you're ready to restore your master branch, the easiest way to do this would be to git reset --hard upstream/master (assuming you set up the original remote name as upstream, as most people here seem to, if the original source repo remote you gave some other name, obviously replace that), which will update your local branches. Then git push -f to force push that update the master on your origin branch. Thenceforth remember that commiting directly to master is a really bad idea.

[rogancarr]

You commited directly into master branch on your own fork about a week ago, somehow didn't catch it, pushed that cahnge, and you've been doing pulls ever since.

Once you're ready to restore your master branch, the easiest way to do this would be to git reset --hard upstream/master (assuming you set up the original remote name as upstream, as most people here seem to, if the original source repo remote you gave some other name, obviously replace that), which will update your local branches. Then git push -f to force push that update the master on your origin branch. Thenceforth remember that commiting directly to master is a really bad idea.

Yes. I fixed that a couple days ago, but this PR still has the remnants :)

[TomFinley]

Thank you @rogancarr , this makes me a bit more comfortable.

[singlis]