This message was deleted.

Hello, Yes, it will choose the hyperparameters that minimize the validation loss. So the use of a validation set combined with early stopping should prevent overfitting. With the validation set, we are evaluating the model's abilitiy to make predictions from training on previous data. I hope this helps!

thanks Marco, ill make sure to use early stopping to prevent overfitting

I experienced a similar issue. Early stopping functioned correctly when I utilized the TFT model. However, when I switched to using AutoTFT, early stopping didn't seem to work as expected. The validation loss continued to increase beyond the designated early_stop_patience_steps.

yah I dont think it works with AutoTFT

Neither works for me with the AutoModels of nf. The val loss increases and increases check by check and yet early stopping is not triggered.

@José Morales FYI

Thanks for reporting this issue. We will take a look at it and fix it!

@nickeleres @Steffen @Rafiq Darwis Mohammad are you using ray or optuna?

Ray @José Morales

Thanks! I'm able to reproduce the issue. I'll keep investigating.

ok lmk when ray is fixed 🙂

Found the issue. The sad part is that I had already seen it a couple of months ago but forgot about it. I'll work on a fix and we're planning on making another release soon so it'll be available probably next week

brilliant thanks Jose

It's on PyPI now

brilliant thank you Jose