Search code examples
glmh2odriverless-ai

Explicitly make GLM model on h2o driverless AI

I have done a few experiments on h2o driverless AI. Was curious to know whether we can explicitly ask the experiment to train using GLM.

Or it will always make model the way it does?

Solution

  • [ Driverless AI is a brand new product, so it's evolving very fast. This answer is for version 1.0.5 of Driverless AI, but could quickly become out of date. ]

    Driverless AI uses (preferably GPU-based) xgboost tree models inside for evaluating variable importance during feature engineering, and for building a final model to make predictions. But what you can do, after the experiment progress bar shows 100% complete, is download the transformed training and test data as .csv files. These transformed .csv files include the engineered features as new columns, and can be used as inputs to H2O-3's GLM, for example.

    Adding more kinds of models definitely makes sense for the future, though.