Offline conversation with @TomFinley

Ranking can be done, pretty straightforward. We just need to make sure that we handle GroupId and an option SamplingKeyColumn together. (e.g. Imagine identifying an author by sentence. We'd want to have groups of sentences for ranking, and make sure that the set of authors in a hold-out set were different from the set in the training set)

Also, TrainTest should similarly be modified to support both GroupId and SamplingKeyColumn. Right now, it only allows one SamplingKeyColumn and supposes that the end user understands that GroupId must be a SamplingKeyColumn.

CrossValidation not available for Ranking, TrainTest not suitable for Ranking #2685

Description

rogancarropenedon Feb 21, 2019

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

rogancarr
openedon Feb 21, 2019