NHITS

NHITS (Neural Hierarchical Interpolation For Time Series) is a deep learning model that uses multi-rate data sampling and hierarchical interpolation to efficiently capture both short-term and long-term temporal patterns. It is particularly suited for long-horizon forecasting.

Training parameters

Patience

Number of validation iterations before early stopping, -1 for unlimited.

Max steps

Batch size

The size of the batches to be used for training and prediction.

Random state

Using a fixed random seed allows for reproducible result.

Model parameters