Add Generalized Additive Models (GAMs) to H2O
The following subtasks will make it to the major release in end of march:
PUBDEV-7178: Generate basic GAM Schema and Algorithm
PUBDEV-7180: added modelmetrics
PUBDEV-7181: Python API
PUBDEV-7182: R API
After digging around the literature, I have come up with the following plan and subtasks:
Generate basic GAM Schema and Algorithm supporting regression (PUBDEV-7178);
Generate GAM dataset and implement Java unit test (PUBDEV-7178);
Generate ModelMetric and implement model metrics calculation (PUBDEV-7180);
Generate Python API o support GAM and implement Python unit test (PUBDEV-7181);
Generate R API to support GAM and implement R unit test (PUBDEV-7182);
Generate Python API to support GAM model metrics and implement Python unit test (PUBDEV-7183);
Generate R API to support GAM model metrics and implement R unit test (PUBDEV-7184);
Generate GAM Mojo (PUBDEV-7185);
Note that task 4 and 5 can start as soon as 1 is completed. Tasks 6, 7 can start as soon as task 3 is completed. Task 8 can start as soon as task 1 is completed.
In addition, we will want to do the following as well:
Add support for more smoothers;
Add methods for choosing knots;
Add support for other families and links: binomial, multinomial, gamma, ….. Check with R and see what families they support in mgcv toolbox.
I will add the appropriate JIRAs as we go along.