Alexander März

@hashbangCoder The error *Exception: Reached maximum number of idle transformation calls.* usually comes from if many of your time series are shorter than your forecasting horizon. You might want to...

@hashbangCoder Are you using ```MultivariateGrouper``` and ```grouper_train``` for grouping the data? For an example on how to use it, see [here](https://github.com/zalandoresearch/pytorch-ts/blob/master/examples/Multivariate-Flow-Solar.ipynb)

One way to provide insights into feature importance of the model is to use attention-weights, as used by the [Temporal-Fusion-Transformer](https://github.com/awslabs/gluon-ts/tree/master/src/gluonts/model/tft). @mbohlkeschneider Any plans to provide some convenience functions in gluonts...

@neverfox Thanks for the PR. Before merging, can you please - include `try` also for the `mean` and `variance` in case one or all of them are not available -...

- Yes, having a `get_moments` function that has the `fitted distribution` as an argument within the `DistributionClass` makes sense - If one of the moments is not implemented, returning `None`...

I was wondering that if the variance or mode are not available in the distribution class, we can also infer them from the predicted samples. Can you add this maybe.

> Sure. It might make sense to make that behavior explicit so the user knows what they're getting? Yes, please make sure to adjust the function description to make it...

Thank you for implementing the changes! I am currently in the process of completing the unit tests in the upcoming weeks. Once they are finalized, you can modify them accordingly...

can you please also add unit-tests for your function, thanks.

@mschrader15 Thanks for opening the PR. Can you please add a unit-test alongside with the code changes. Thanks!