IBM PatchTST-FM pretrained foundation model for time series forecasting.

Uses the zero-shot pretrained PatchTST-FM (ibm-research/patchtst-fm-r1) from IBM, a ~260M parameter model that achieves state-of-the-art results on GIFT-Eval. Requires granite-tsfm from the patchtst-fm branch.

PatchtstFmHyperparams

patchtst_fm_model.PatchtstFmHyperparams

Stochastic output shape (pretrained weights fixed; no training grid).

Attributes

Attribute	Type	Default	Description
stochastic_samples	int	100	How many quantiles to subsample into pseudo-samples at predict time.

PatchtstFmModel

patchtst_fm_model.PatchtstFmModel

IBM PatchTST-FM pretrained foundation model wrapper.

Zero-shot stochastic forecasting via the granite-tsfm package (patchtst-fm branch). PatchTST-FM outputs quantile forecasts which are converted to pseudo-samples.

__init__(self, params: Dict[str, Any], settings: Dict[str, Any])

Parameter	Type	Default	Description
params	Dict[str, Any]	-	(undocumented)
settings	Dict[str, Any]	-	(undocumented)

train(self, y_context: np.ndarray, y_target: np.ndarray, timestamps_context: np.ndarray, timestamps_target: np.ndarray, x_context: Optional[np.ndarray] = None, x_target: Optional[np.ndarray] = None, **kwargs) -> PatchtstFmModel

Load pretrained PatchTST-FM (no training required for foundation models).

Parameters:

Parameter	Type	Default	Description
y_context	np.ndarray	-	(undocumented)
y_target	np.ndarray	-	(undocumented)
timestamps_context	np.ndarray	-	(undocumented)
timestamps_target	np.ndarray	-	(undocumented)
x_context	Optional[np.ndarray]	None	(undocumented)
x_target	Optional[np.ndarray]	None	(undocumented)
**kwargs		-	(undocumented)

Returns: PatchtstFmModel (Self).

predict(self, y_context: np.ndarray, timestamps_context: np.ndarray, timestamps_target: np.ndarray, x_context: Optional[np.ndarray] = None, x_target: Optional[np.ndarray] = None, **kwargs) -> np.ndarray

Parameters:

Parameter	Type	Default	Description
y_context	np.ndarray	-	(undocumented)
timestamps_context	np.ndarray	-	(undocumented)
timestamps_target	np.ndarray	-	(undocumented)
x_context	Optional[np.ndarray]	None	(undocumented)
x_target	Optional[np.ndarray]	None	(undocumented)
**kwargs		-	(undocumented)

Returns: np.ndarray (The predicted samples, shape (num_samples, forecast_horizon, num_targets)). Raises: ValueError if PatchtstFmModel is not fitted (i.e., train() has not been called).