continue_training parameter in fit() does not exist in the main branch but is documented in the official site

[anandaheino]

While me and @heinoso were trying to use the continue_training parameter in the forecaster.py/fit() method as described in the official documentation we encountered an issue.

According to the documentation, this parameter is meant to allow continued training from the last checkpoint. However, when I try to fit the model again using this parameter, I receive the following error:

RuntimeError: Model has been fitted already. Please initialize a new model to fit again.

Inspecting the source code in the main branch, the fit() method does not include the continue_training parameter. The only parameters available are:

neural_prophet/neuralprophet/forecaster.py

Lines 970 to 989 in 5e6b231

	def fit(
	self,
	df: pd.DataFrame,
	freq: str = "auto",
	validation_df: Optional[pd.DataFrame] = None,
	epochs: Optional[int] = None,
	batch_size: Optional[int] = None,
	learning_rate: Optional[float] = None,
	early_stopping: bool = False,
	minimal: bool = False,
	metrics: Optional[np_types.CollectMetricsMode] = None,
	metrics_log_dir: Optional[str] = None,
	progress: Optional[str] = "bar",
	checkpointing: bool = False,
	num_workers: int = 0,
	deterministic: bool = False,
	scheduler: Optional[Union[str, Type[torch.optim.lr_scheduler.LRScheduler]]] = None,
	scheduler_args: Optional[dict] = None,
	trainer_config: Optional[dict] = None,
	):

After further investigation, I found that the continue_training parameter is present only in the train-continue branch, which depends on the custom-lr-scheduler branch.

Suggestions

• Include the continue_training parameter in the fit() method in the main branch.
• Ensure the documentation aligns with the main branch to avoid confusion for users.
• If this feature is still under development, clarify in the documentation that it is available only in specific branches/versions.

Final Comentary

We really enjoy using NeuralProphet model, it gives really great results and I would love to be able to refit the model, just like in other models (e.g. Prophet, XGBoost, LightGBM, etc). Keep up with the great work :)