Early stopping support for sklearn Pipelines

ray-project / tune-sklearn

A drop-in replacement for Scikit-Learn’s GridSearchCV / RandomizedSearchCV -- but with cutting edge hyperparameter tuning techniques.

https://docs.ray.io/en/master/tune/api_docs/sklearn.html

Apache License 2.0

465 stars 52 forks source link

Early stopping support for sklearn Pipelines #103

Closed Yard1 closed 4 years ago

Yard1 commented 4 years ago

Allows to use early stopping on sklearn Pipelines. The last step (almost always the estimator) will be treated to the same early stopping checks as non-Pipeline estimators.

Adds a new param to TuneSearchCV and TuneGridSearchCV, pipeline_detection (open to better name suggestions). If True, will handle early stopping if the estimator is a Pipeline.
Added a new _Trainable child class, _PipelineTrainable, handling early stopping for Pipelines.
Added an example.

@richardliaw

richardliaw commented 4 years ago

last comment - could you add a quick unit test or two in test_gridsearch.py and/or test_randomizedsearch?

Looks good and will merge after above comments + a unit test. Thanks a bunch @Yard1 !

richardliaw commented 4 years ago

looks like a couple tests are failing - let me know if you run into any trouble fixing those!

Yard1 commented 4 years ago

@richardliaw everything aside one Mac job passes.

richardliaw commented 4 years ago

@Yard1 awesome!