Streamline "freq", "time_granularity" naming

[mbohlkeschneider]

Find a naming that is consistent and change it throughout the code for freq/time_granularity.

[geoalgo]

Given that Pandas use 'freq' (link), this is probably the best option.

[mbohlkeschneider]

I agree with that. We should make this distinction clear somewhere in the documentation, though, that this is not confused with seasonality frequency.

[jaheba]

I'm currently looking into the shell (essentially integrating with SageMaker) and noticed this too.

In SageMaker DeepAR we use time_freq and I think there is an argument to be made that we should aim for consistency.

[mbohlkeschneider]

Consistency is a good argument. But consistency with SageMaker might not necessarily be the right thing in a library. Pandas uses freq for generating time series Series objects and I think it would be good to stick to that.

[jaheba]

I'm fine with either, but at least we should be consistent within GluonTS.

Do we have a set of required hyper-parameters? Besides freq prediction_length comes to mind.

[geoalgo]

Yes, freq and prediction_length are the only two required hyper-parameters.

Pandas is much more frequent than Sagemaker, so I also freq will be nicer.

[lostella]

addressed in #172 and #173