Repository navigation
Update or deprecate AutoML 1.x APIs #6532
Description
Activity
- changed the title
[-]Update or deprecate AutoML v1 APIs[/-][+]Update or deprecate AutoML 1.x APIs[/+]on Jan 4, 2023 @luisquintanilla
I'm also in favor of proposal 1. The following AutoML v1 APIs have already been updated to Flaml implementation- binary classification
- multi classification
- regression
And the rest of scenarios are WIP
- recommendation
- image classification
Reacted by Luis Quintanilla and Simon MattesGreat! Leaving this issue open to track progress. Thanks!
I'd also promote option 1, because deprecation would seriously impede existing investments into ML.net for a lot of clients.
Reacted by Luis Quintanilla, Simon Mattes and Antti "Andy" TörrönenAlso in favor of no.1 to avoid rewrites, if possible. Perhaps it could be considerd if a new FLAMLSweeper could be added so the old code structure could stay the same, and potentially keep the old sweepers in codebase. It would help us who have customized the AutoML code and we could merge the new codebase without big rewrites.
Reacted by Luis Quintanilla and Simon Mattes@LittleLittleCloud are we okay to close this issue?
Yes let's close it
- locked and limited conversation to collaborators
on Jan 8, 2024
Metadata
Metadata
Assignees
Labels
No labels
Problem
Currently both ML.NET 1.x and ML.NET 2.x AutoML APIs are available to users. Although users can still run old samples, the differences in the implementations cause confusion.
Proposal(s):
Personally I'd be in favor of 1 since it would not break old samples while at the same time providing an even higher layer of abstraction on top of the Experiment API. However, in the event the breaking changes are significant, I'd be in favor of 2.
User evidence