Additional information about feature engineering in H2O or Time-series

[hoanglam-novobi]

Hi there,
I am learning about the platform to use in our project to create a predictive model like sales forecast...
Anyone can help me to answer some question about it:
1/ How does H2O create new features in feature engineering?
2/ What are the primitives (min, max, mean, mode, ... or some special primitives to solve some problem in time-series)?
3/ What is the maximum/minimum value of prediction length in the feature Time-series in H20? From your case study about sale forecasting, the maximum value of prediction length is 39 weeks. Can H2O do with the different interval time in the data like minutes, ..?
4/ For performance, can we have set the number of cores CPU to use in the training process? Is it supported parallel?

Thank you.