search

openml-labs / gama

An automated machine learning tool aimed to facilitate AutoML research.
https://openml-labs.github.io/gama/master/
Apache License 2.0
92 stars 29 forks source link

Merge Master to clustering for new changes in GAMA Backend #181

Open prabhant opened 1 year ago

simonprovost commented 7 months ago

@PGijsbers It would be fantastic to look into this one as well, following #210! I will keep that in mind, and if I have a day or so, I will see if we can create a brand-new PR with the new additions so that Classification Regression and Clustering will all be available with ConfigSpace ☀️ Yet, Clustering will in anyway not help my Ph.D so will have to look into that in my spare time.

PGijsbers commented 7 months ago

I think has already diverged from main quite substantially. It's likely easier and better to look into a re-implementation rather than cleaning this up (especially after #210 is merged). Besides, clustering has a number of difficulties with the AutoML paradigm that GAMA uses (the internal metrics (i.e. they don't use labels) don't transfer that well to performance on external metrics (i.e., ones that do, and thus can evaluate performance based on ground truth). I am not entirely sure if it makes sense to integrate clustering at this point, which is one of the reasons I (and Prabhant) haven't put real effort behind merging this PR.

simonprovost commented 7 months ago

I think has already diverged from main quite substantially. It's likely easier and better to look into a re-implementation rather than cleaning this up (especially after #210 is merged). Besides, clustering has a number of difficulties with the AutoML paradigm that GAMA uses (the internal metrics (i.e. they don't use labels) don't transfer that well to performance on external metrics (i.e., ones that do, and thus can evaluate performance based on ground truth). I am not entirely sure if it makes sense to integrate clustering at this point, which is one of the reasons I (and Prabhant) haven't put real effort behind merging this PR.

Indeed! It makes sense now. Will not focus this then. Yet, when you have time, maybe put a label on the PR to avoid any future contributor ^^

Have a great day,

Cheers,