Lorenzo Stella

@fernandocamargoti did you mention you were running on GPU? In that case this may be related: https://github.com/awslabs/gluon-ts/pull/728

I'm wondering whether this still shows up with MXNet 1.7? (since that fixed the bug concerning #728) @fernandocamargoti not sure you're able to verify this, otherwise we'll try to reproduce...

@fernandocamargoti we still have to reproduce the issue on `master`. I’m trying with the m5 data that you provided, but for debugging purposes it would be nice to have a...

I agree fully on using short imports in the documentation, tests and throughout the package. What would be the benefit of the serde story? Avoid breaking objects when a class...

Still some issue with `dtype`: ``` E AssertionError: The index of the forecast is outside the target index E Index of forecast: DatetimeIndex(['2018-01-08', '2018-01-09', '2018-01-10'], dtype='datetime64[ns]', freq='D') E Index of...

@Vzlentin this is expected behaviour: you can call `as_symbol_block_predictor` as follows, to force hybridization of the prediction network. ```python predictor = estimator.train(train).as_symbol_block_predictor(dataset=train) ``` You can also consider the PyTorch implementation...

@Vzlentin sorry for the late reply. It’s hard to say, but it’s a shape issue. Could you come up with a minimal example reproducing the issue? With fake data it’s...

@eugeneteoh exactly, thanks for the issue: flexibility of the evaluator is something we're currently working on, I'll keep you posted when there are updates

Some general comments: - I like the decorator approach. - There is at least one metric ([MSIS](https://github.com/awslabs/gluonts/blob/d5bf5bf147a52798104f4c254d246ed863a68584/src/gluonts/evaluation/metrics.py#L132-L154)) which is only defined per "test entry" and is not directly the mean/sum...

Where would you apply the decorator? To `GluonEstimator.train` for example? Then it’s not too different from doing it manually there, since it’s just a couple of lines (also not a...