Consistent ordering of (feature, label, groupid, weight) parameters in ML.NET trainer estimators

[abgoswam]

The public API of ML.NET should have a defined ordering for

• feature column name
• label column name
• weight column name
• group id column name

Currently the trainer estimators in ML.NET do not have a defined ordering for the above parameters. We should address this as we finalize the API for V1.0

Here is a proposal for consistent ordering of these parameters for the trainer estimators:

1. unsupervised trainer estimators (e.g. KMeansPlusPlusTrainer)
   (feature , weight)

2. supervised trainer estimators: (e.g. LightGbmBinaryTrainer)
   (label, feature, weight)

3. ranking trainer estimators: (e.g. LightGbmRankingTrainer)
   (label, feature, groupid, weight)

Related to #2177, #2257, #2660

@Ivanidzo4ka @sfilipi @rogancarr @TomFinley @glebuk

[abgoswam]

I verified the above pattern is followed by the trainer estimators. The only exception is FieldAwareFactorizationMachine which differs in 2 ways .

1. it accepts a string[] for featureColumnNames . other trainer estimators just take a string
2. label is the 2nd parameter . in the other trainer estimators label is the 1st argument

Marking as closed.