Have your H2O ever exceeded the default training time budget (1 hour) and you had to stop it? what is your interpretation of this problem?
I am very familiar with this solution and this is happening for the very first time with a multiclass dataset of 20k instances, 40 features and 103 classes. However, surprisingly, when I lower the time budget to 15 mn (for example), h2O returns a model and makes predictions.
Thank you,
Yassine
