Back to models
Forecaster

TiRexForecaster

API referenceSource
Categorical featuresIn-sample predictionsIn-sample prediction intervals

Interface to the TiRex Zero-Shot Forecaster.

This forecaster loads the TiRex model from the tirex-ts package when fit() is called. Instead of training, it takes the given data as context, and predict() uses that context to produce forecasts for the requested future time points. torch is required at runtime and a clear error is raised if unavailable.

Quickstart

python
from sktime.forecasting.tirex import TiRexForecaster

estimator = TiRexForecaster(model='NX-AI/TiRex', device: str='cpu', license_accepted: bool=False)

Tags

Capabilities

  • Categorical features
  • In-sample predictions: Supported
  • In-sample prediction intervals: Supported
  • Prediction intervals: Not supported
  • Missing values: Not supported
  • Exogenous variables: Not supported
  • Multivariate: Not supported

Properties

Needs forecast horizon in fitrequires-fh-in-fit
No
X and y need the same indexX-y-must-have-same-index
Yes

Parameters(3)

modelstr (default = “NX-AI/TiRex”)
“Model identifier to load via the vendored TiRex loader”
device{“cpu”, “cuda”, …}, default=”cpu”

Compute device used by the underlying TiRex model. "auto" selects CUDA, then MPS, then CPU.

license_acceptedbool, default=False
Whether the user accepts the license terms of TiRex. Must be set to True to use the model.

References

[1] TiRex: https://github.com/NX-AI/tirex