Nixtla's Open Source Time Series Ecosystem.

Nixtla Community

when training a NHIST model via `nf.core.NeuralForecast.fit()`, is there a reason we cannot set the size of the test set?

Hello! The `fit` method expects data for training only, not the entire dataset. So, it's a bit like in sklearn, we do a train/test split first, then we past the training data to the `fit` method.

You can specify a validation size if you want to early stop, using the `val_size` parameter.

<@U06HEUYHE05> follow up question, how am I supposed to get the training loss so I can plot it to see if overfitting or undertraining is happening?

nvm I found it. It should be an example in the docs

```from pytorch_lightning.loggers import TensorBoardLogger

tb_logger = TensorBoardLogger(save_dir="tb_logs", name="tb_logs")
nf = NeuralForecast(models=[NHITS(..., logger=tb_logger)])
nf.fit(...)```
Then in another shell:

```tensorboard --logdir ./tb_logs```