Nils Kiele

Thanks for your thoughts. 👍 In my head, the time axis is always 1 but being able to specify `time_axis` would make things more general, true. I'm not sure your...

While applying the new `Evaluator` to more data (first 400 entries) of the electricity dataset, I found some fundamental problems I haven't given much thought before: * **Division by zero**...

Update: My concern that “time series of different length don’t currently work” was resolved by not keeping the input time series as part of the data dict. Instead, we now...

After adding a rough comparison of the old and new approach (`comparison_to_old_approach.py`) to see if the results make sense, I found that: - (Weighted) Quantile Loss is [multiplied by two](https://github.com/awslabs/gluonts/blob/5ee4e451e346076fbcdf66f8dc5d30b1008216f1/src/gluonts/evaluation/metrics.py#L75-L81)...

Now, there's a test comparing the results of the former `Evaluator` class with the new evaluation approach., using the deterministic `SeasonalNaivePredictor` to forecast. Invalid value handling (`NaN`s and `inf`s) was...

There's an open PR (#2338), trying to unify plotting utilities. [This function](https://github.com/awslabs/gluonts/blob/7c44b3c456fa6f937273616b12da646deb4b2c31/src/gluonts/lab/viz.py#L155-L249) allows plotting multivariate forecasts, either in subplots (`use_subplots=True`) or all in one plot (which gets messy quickly, depending...