Open why362 opened 2 years ago
I did not dive that deep into the issue, but I also encountered it while using Google Colab. Then, I installed the library locally, and the issue was gone. Also, you can see here that the loss functions are not actually gone from the codebase: https://github.com/jdb78/pytorch-forecasting/blob/master/pytorch_forecasting/metrics/distributions.py#L43.
Still an issue has anyone resolved this - on Google Colab?
I was trying to build a DeepAR model to solve a multivariate time series forcasting problem according to the tutorial in the offical website, https://pytorch-forecasting.readthedocs.io/en/latest/tutorials/deepar.html, only to find that the MultivariateNormalDistributionLoss had been removed from the source code. Why remove it? And what should I do to tackle this problem using DeepAR?