I looked for a generalized linear model implementation with regularization. I found that glmnet does not allow custom link function. However, h2o takes link function type as a parameter. Is it possible to define and use a custom link function under a family (optimization problem is same) in h2o?
Asked
Active
Viewed 118 times
1 Answers
0
The answer depends on how badly you need it...
No: you have to use one of the built-in link functions, and cannot pass your own through the API.
Yes: it is open-source, so you could hack in your custom link function to the Java back-end, give it a name, recompile, and then it should be available from the API.
I poked around in the Java source a bit, hoping to be able to give you a single link in a single file where you add your code. But it looks more complicated than that, I think it is a non-trivial project. (If you have more budget than time, becoming a paying h2o.ai customer, and making a feature request is likely to be the quickest way to get it done!)

Darren Cook
- 27,837
- 13
- 117
- 217