You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I figured out what is causing the error - when building the model, I use a custom quantile range of [0.1, 0.5, 0.9]. When I leave this parameter out and use the default QuantileRegression, the loader works.
Describe the bug
I have trained a TFT model with Quantile loss and I cannot load the checkpoint that was saved during best epoch for evaluation.
To Reproduce
I let the model train like so and then I want to load the best checkpoint for evaluation:
but that leads to the following error:
TypeError: cannot assign 'darts.utils.likelihood_models.QuantileRegression' as child module 'criterion' (torch.nn.Module or None expected)
Can anyone please suggest a solution?
System (please complete the following information):
The text was updated successfully, but these errors were encountered: