Add weights hyperparemeter to EnsembleForecaster API

Is your feature request related to a problem? Please describe. EnsembleForecaster does not currently accept weights. With #1136 and #1139 being worked on it would be nice to optionally allow users to pass weights to EnsembleForecaster.

This would give users an out-of-the-box way to use ainverse error based weighting strategy in forecasting ensembles. Users could also supply other types of weights (information criterion, etc).

Describe the solution you’d like An update to the EnsembleForecaster API to allow weights to be passed (number weights should equal number of forecasters passed).

This would require altering the __init__ to store the weights, adding a check (in _fit?), and then updating _predict to use versions of the aggfunc that allow weights (for examples of this see use of _weighted_percentile, _weighted_geometric_mean, and np.average in performance_metrics_functions.py).

Describe alternatives you’ve considered N/A

Additional context N/A

[RNKuhns]

@GuzalBulatova that works for me. @aiwalter and I have been talking about some options for the functionality. Maybe the three of us can find a quick time to talk.

After digging into the online ensemblers per @fkiraly’s comment above. I think that the current differences between that and EnsembleForecaster are currently that the online ensemblers use weights and update those weights (the online part).

I think that it would be reasonable if the only difference was that the online ensembles update their weights (EnsembleForecaster would fix its weights based on the new weights hyperparameter).

[GuzalBulatova]

Hi @RNKuhns I’d like to work on this 🚀 It’s closely related to #1082 that I’m working on now.