2

I'm wondering how to standardize features when using h2o's AutoML with deep learning and GLM algorithms.

Seems it is supported to deep learning and GLM models (https://docs.h2o.ai/h2o/latest-stable/h2o-docs/data-science/algo-params/standardize.html), but in h2o.automl it does not accept the standardize = TRUE argument.

My questions are:

  1. Does autoML automatically scales (i.e. standardizes) the features when deeplearning or GLM algorithm is used?

If true, does it automatically standardize also when I predict on new test data?

  1. If 1) is not true, is there a built-in h2o function that achieves this so that I can do it manually? What's the recommended workflow for this with AutoML?
Matthew Son
  • 1,109
  • 8
  • 27

1 Answers1

2

  1. Yes, H2O AutoML uses most of the hyperparameter defaults in GLM and Deep Learning, and both of those default to standardize = TRUE.

  2. In H2O, every transformation that happens in training will happen at predict time, so you don't need to worry.

Erin LeDell
  • 8,704
  • 1
  • 19
  • 35
  • @LeDell Thank you very much. Does standardize = TRUE in Deep Learning, GLM, ..., standardize the target variable altogether, or only features? – Matthew Son Apr 28 '23 at 16:00
  • Linking your answered question here: https://stackoverflow.com/questions/76131318/h2o-variable-standardization – Erin LeDell Apr 28 '23 at 21:23