search

linkedin / photon-ml

A scalable machine learning library on Apache Spark
Other
790 stars 185 forks source link

Examples for hyperparameter tuning #471

Closed qingquansong closed 4 years ago

qingquansong commented 4 years ago

Can we add some examples for using RandomSearch and BayesianOptimization functions to tune the GLMix models? Thanks!

cmjiang commented 4 years ago

Hi @qingquansong, BayesianOptimization is no longer supported in Photon ML at this moment. The PR https://github.com/linkedin/photon-ml/pull/454 aims at removing hyperparameter autotuning in this repo. Grid search is still supported as the default tuning method.

qingquansong commented 4 years ago

Thanks for the reply! Another examples for grid search? I saw that we can set a list of regularization parameter directly. But is that all we can tune?

cmjiang commented 4 years ago

Yes, that list is about a grid search. Once you specify a list of the regularization, a list of the models in the same length will be trained and you can find the best model in the output.

yunboouyang commented 4 years ago

Just want to mention that you can specify regularization weights for each component of GLMix, such that [1, 2] for a fixed effects model, [1, 10] for a per-member model and [1, 100] for a per-job model. In this case photon-ml will do cross products and train 8 models in total.

qingquansong commented 4 years ago

Just want to mention that you can specify regularization weights for each component of GLMix, such that [1, 2] for a fixed effects model, [1, 10] for a per-member model and [1, 100] for a per-job model. In this case photon-ml will do cross products and train 8 models in total.

Thanks for the clarification!