Attribute Error for pytorch_forecasting models

Question

I'm currently trying to set up a TemporalFusionTransformer from the pytorch_forecasting library in Google Colab, following the exact approach that is stated here.

For some reason, I always get "AttributeError: 'tuple' object has no attribute 'items'" when trying to set up a model.

Even if I just try initializing a default model in the Baseline class, I get the same error (see screenshot). Error message for Baseline model Maybe someone knows how to solve this issue here. Thanks in advance.

What I tried so far:

• converted my data into a TimeSeriesDataSet object
• used TemporalFusionTransformer.from_dataset(...) providing my data as well as some training parameters
• tried the same without providing the training parameters
• tried setting up the TFT with and without training parameters (i.e. TemporalFusionTransformer(...))
• tried the same with the Baseline model (Baseline())

result was an Attribute error in all cases

Answer 1

pytorch_lightning had an update (https://pypi.org/project/pytorch-lightning/#history). Downgrading to version 1.9.0 worked for me.

you can use pip install pytorch_lightning==1.9.0