I was wondering which package to use between pytorch forecasting (https://pytorch-forecasting.readthedocs.io/en/stable/) or darts (https://unit8co.github.io/darts/). I have been trying both, it looks like darts is more sklearn-like in its writing and style and pytorch forescasting uses different data classes.
Any comment comparing the two would be welcome.
I don't know if some of you might have carried out a performance comparison between both libraries.
Thanks in advance!
I think one of the biggest advantage of darts is its Timeseries Object which is very pandas-like and very intuitive when you are familiar with sklearn. However, I also do see the advantage that pytorch-forecasting dealt with categorical data "better" (easier) and it takes a steeper learning curve to understand pytorch-forecasting. I would say pytorch-forecasting sometimes outperform darts using the same model.
If helpful, I have just made a pipeline for a Temporal Fusion Transformer model in Darts and am considering rebuilding it in PytorchForecasting for the following reasons:
As mentioned above, the dataset handling in Darts is pretty good and they have abstracted away the Pytorch dataloader
