Add a `model.supports_sampled_predictions` attribute

[felixdivo]

Is your feature request related to a current problem? Please describe. It would be great to get the properties listed in the nice table in the README also from code. This helps with automated evaluation of multiple models. Additionally/Alternatively, having a model.supports_probabilisitc_predictions would be nice, where it would be equal to model.supports_sampled_predictions or model.supports_likelihood_parameter_prediction.

Describe proposed solution I'd like to be able to do something along these lines

my_model = ModelClass(some="properties")
print(f"can sample forecasts: {my_model.supports_sampled_predictions}")
print(f"can estimate distribution parameters: {my_model.supports_likelihood_parameter_prediction}")
print(f"can make probabilistic forecasts: {my_model.supports_probabilisitc_predictions}")

Describe potential alternatives You make a list of the models that can do this by yourself. It works, but is annoying.

Also, one might discuss the distinction between supports (the implementation can do it, in principle) and uses (the parameters that are passed enabled it).

Additional context Suggestions for other namings of these attributes are more than welcome. :)

[dennisbader]

This is a good point @felixdivo . Currently we have this as a private model method ForecastingModel._is_probabilistic().

I agree that we should make this public and similar to the other support properties.

Here are some thoughts/points to keep in mind:

• Currently, all models that support probabilistic predictions, support the sampled predictions.
• Not all of these models support the likelihood_parameter predictions
• Global models are either probabilistic (when they use a likelihood) or deterministic (when they don't)
• On the other hand, the local models are deterministic but they can also support probabilistic predictions (ususally some kind of mc simulation).

Maybe converting the private method ForecastingModel._is_probabilistic() into a public property ForecastingModel.supports_probabilistic_prediction would do the trick?

[felixdivo]

Sounds reasonable! Would you like me to propose a PR for it?

[dennisbader]

Yes, go ahead @felixdivo, thanks!

[madtoinou]

Fixed by #2269.