Support for recursive multistep forecasts

[jzicker]

I'm not sure if this will be a question or request for additional functionality. In general support for machine learning multistep forecasts would be extremely beneficial. Currently the recursive functionality somewhat supports. I may not be using it correctly however as a few steps in the flow are a bit bumpy.

1. I want to split the data into test and train and then smoothly calculate accuracy and plot the forecast over the recursive horizon for the test set. Currently I split the data and then use the train portion to extend into future and then have to insert actuals into the future frame to get this to work. Maybe I'm doing something incorrectly
2. Next step is to perform cross validation using a recursive multistep forecast and calculate accuracy two ways. The first is to average the mape of all the slices together and the second is to provide a mape for each time step in the horizon. I think of this as having a mape value for each step in a recursive forecast and then compute mean and standard deviation using all of the step n from the slices.
3. Of course would like for this to work for panel and ensemble functionality as well.

Thank you.