Artem Budishchev comments

Results 6 comments of


                                            Artem Budishchev

TemporalFusionTransformer predicting with mode "raw" results in RuntimeError: Sizes of tensors must match except in dimension 0

Not sure if it's the same issue, but we're facing the same problem when running the latest version of the package. What we noticed is that if you set `return_y...

Temporal Fusion Transformer gives constant prediction

@valentinfrlch for clarity, are you saying that _disabling_ GroupNormalizer is what did the trick for you?

RuntimeError: Sizes of tensors must match except in dimension 1

Please see my comment here - https://github.com/jdb78/pytorch-forecasting/issues/449#issuecomment-1649288069. If you don't need the ys (it's easy to format them yourself), then setting `return_y = False` fixes the issue. @hippotilt thanks! I...

% values were found to be NA or infinite (even after encoding)

You probably have some negative values in your dataset, which makes softplus transformation produce negative infinities. I would probably just shift the whole feature so that there are no more...

Using parsestream blocks until fully read

Sorry for reviving an old issue, but it would be great if something similar (exactly the same?) as what @Toilal has added in the fork was added upstream!

TFT prediction at inference time for unseen entity key fails

Actually in our experiments GroupNormalizer tends to perform better than the EncoderNormalizer. Unfortunately, this means that we cannot do inference for the unseen groups. One thing to consider though is...