Implementation:Sktime Pytorch forecasting Trainer Fit

Knowledge Sources	PyTorch Lightning Trainer.fit pytorch-forecasting
Domains	Deep_Learning, Training, Optimization
Last Updated	2026-02-08 07:00 GMT

Overview

Wrapper documentation for the PyTorch Lightning Trainer.fit method as used in pytorch-forecasting workflows to execute model training.

Description

The Trainer.fit method executes the full training loop for a pytorch-forecasting model. It iterates over epochs, running training steps (via BaseModel.training_step at L751-808) and validation steps. For each training batch, the model computes the forward pass, calculates loss (QuantileLoss, NormalDistributionLoss, or MASE depending on model), applies gradient clipping, and updates parameters. Callbacks like EarlyStopping and LearningRateMonitor execute at appropriate hooks. The method is external to pytorch-forecasting (from Lightning) but integrates deeply with the library's BaseModel training infrastructure.

Usage

Call after all setup is complete: Trainer configured, model instantiated, DataLoaders created, and optionally learning rate found. This is the final training step in all four workflows.

Code Reference

Source Location

Repository: External — pytorch-lightning
File: lightning/pytorch/trainer/trainer.py
Related internal code: pytorch_forecasting/models/base/_base_model.py L751-808 (training_step)

Signature

class Trainer:
    def fit(
        self,
        model: LightningModule,
        train_dataloaders: DataLoader | LightningDataModule | None = None,
        val_dataloaders: DataLoader | None = None,
        datamodule: LightningDataModule | None = None,
        ckpt_path: str | None = None,
    ) -> None:
        """
        Runs the full optimization routine.

        Args:
            model: Model to fit.
            train_dataloaders: Training DataLoader(s).
            val_dataloaders: Validation DataLoader(s).
            datamodule: Optional LightningDataModule.
            ckpt_path: Path to checkpoint to resume from.
        """

Import

import lightning.pytorch as pl
# trainer = pl.Trainer(...)
# trainer.fit(model, train_dataloaders, val_dataloaders)

External Reference

PyTorch Lightning Trainer.fit Documentation

I/O Contract

Inputs

Name	Type	Required	Description
model	LightningModule	Yes	pytorch-forecasting model (TFT, DeepAR, NBeats)
train_dataloaders	DataLoader	Yes	Training DataLoader from TimeSeriesDataSet.to_dataloader()
val_dataloaders	DataLoader	No	Validation DataLoader
ckpt_path	str	No	Path to resume training from a checkpoint

Outputs

Name	Type	Description
(side effect)	None	Model weights are updated in-place; checkpoints saved to disk; logs written

Usage Examples

Standard Model Training

# After trainer, model, and dataloaders are configured:
trainer.fit(
    model,
    train_dataloaders=train_dataloader,
    val_dataloaders=val_dataloader,
)

# Access best model path after training
best_model_path = trainer.checkpoint_callback.best_model_path
print(f"Best model saved at: {best_model_path}")

Related Pages

Implements Principle

Principle:Sktime_Pytorch_forecasting_Model_Training

Requires Environment

Environment:Sktime_Pytorch_forecasting_Core_Python_Dependencies

Uses Heuristic

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment