Is it possible to return estimated parameters along with forecasts?

[deejlucas]

I am working on a DeepAR model with a zero-inflated distribution. I am interested in knowing how many forecasts are zero due to the zero-generating process and how many are zero because of the other distribution. In this use case, there is a theoretical zero-generating process and some thought that zero forecasts by the different processes should be handled differently by the business.

Is it possible to return the distribution parameters along with the forecasts?

If not, any tips on the most direct path to making this happen? Digging into the source code, it looks like the distr object in gluonts>model>deepar>_network.py L607 has an array of distribution parameters. I'm assuming I could work my way up through the stack from here and have these functions return multiple outputs, but I'm not sure how much trouble I'm making for myself if I do that.