For short historical data set, how to setup the hyperparameters to yield good forecast? And is DeepAr yield good forecast with short historical data sets?

[Msaleh87]

I have a dataset of 3 months, 15 min intervals of data points. I resampled it to five. The data has somewhat weekly seasonality, it represents the energy demand of a facility. What are the hyperparameters that I should focus on?

I used DeepAr with year-long historical data sets and the results were good.