[Meta Issue] Changing defaults

[najeeb-kazmi]

First of all, is this a breaking change? @terrajobst suggested it is an acceptable breaking change #2305 (comment), @eerhardt commented that is was decided that changing defaults is acceptable #2305 (comment)

• Change the following defaults based on results from @justinormont:

  • Update default n-gram length for Text Transform to match default text recipe #2870 FeatirizeText default n-grams lengths to match default text recipe:

    • word: 1 -> 2
    • char: no change
    • @justinormont correct me if I am wrong: was the default weighting TF-IDF in TLC? It is currently TF

  • Change default # of iterations in Averaged Perceptron to 10 #2305 AveragedPerceptron default iterations from 1 to 10

• RowGroupColumnName of ranking trainers options class defaults to null #3365 RowGroupColumnName in FastTree and LightGbm ranking trainer Options classes: null -> 'GroupId'

  • This won't actually break anything since a ranking trainer would throw anyway if there was no group ID column specified RowGroupColumnName of ranking trainers options class defaults to null #3365 (comment)
  • This would make it consistent with the API for the rankers without the Options class, where the default is 'GroupId'
  • They should be consistent, at least. Either null everywhere or 'GroupId' everywhere.

cc: @harishsk @justinormont @gvashishtha @eerhardt

[najeeb-kazmi]

Assigning @gvashishtha for triage.

[justinormont]

@justinormont correct me if I am wrong: was the default weighting TF-IDF in TLC? It is currently TF

The default text recipe in TLC uses TF. I believe the current default in ML․NET is also TF, so no changes needed for that.

[eerhardt]

Changing a default parameter value is an acceptable change. Just be sure you still support the old default value - for example if the old default was ‘null’ and you change it to a non-null default, you still should support ‘null’ values.

[michaelgsharp]

With the merging of PRs #5248, #5258, and #5290, this issue has been resolved. Closing this issue.