No native Train-Test-Validate cross fold capability in deep learning Autoencoder

Description

When building an autoencoder deep learning model, I ran across what may be a bug or future feature. I have a somewhat small data set, (50000 obs, 4 categorical features) with categorical features with many levels. I need to break this data set into a test and train subset, ensuring that each feature of the one-hot encoded data set remains in both the training and test data subsets. I attempted to specify the `nfolds` parameter in h2o and obtained a messaging stating that this capability was not available while using autoencoders. Any chance on adding this capability to a future version of h2o?

Assignee

Unassigned

Reporter

Spencer Butt

Priority

Minor
Configure