WindFMForecaster

class WindFMForecaster(model_path='NeoQuasar/WindFM', tokenizer_path='NeoQuasar/WindFM-Tokenizer', device='cpu', columns=None, freq='1h', start='2000-01-01', clip=5.0, predict_kwargs=None, deterministic=False)[source]

WindFM zero-shot forecaster for wind power data.

This forecaster wraps WindFM [1], a foundation model for wind power forecasting [2], through the sktime forecasting interface. This implementation is inference-only and uses the upstream WindFMPredictor preprocessing and autoregressive inference path internally.

Parameters:

model_pathstr, default=”NeoQuasar/WindFM”: Hugging Face repository identifier or local path for the WindFM model. The default is the WindFM checkpoint [3]. Other released checkpoints include WindFM-robust [4].
tokenizer_pathstr, default=”NeoQuasar/WindFM-Tokenizer”: Hugging Face repository identifier or local path for the WindFM tokenizer. The default is the WindFM-Tokenizer checkpoint [5]. The released WindFM-Tokenizer-robust is also available [6].
devicestr, default=”cpu”: Device used for model and tokenizer inference.
columnslist of str or None, default=None: Optional mapping from columns in X to WindFM weather covariates. If provided, it must contain five entries ordered as "wind_speed", "wind_direction", "density", "temperature", and "pressure". If None, X must contain the literal WindFM covariate names.
freqstr or pandas offset, default=”1h”: Frequency used to synthesize timestamps for WindFM when the training index is not a pd.PeriodIndex or pd.DatetimeIndex. The default is one hour, matching the example data distributed by WindFM.
startstr or pd.Timestamp, default=”2000-01-01”: Start timestamp used with freq to synthesize timestamps for WindFM when the training index is not datetime-like. The default date is arbitrary; it provides deterministic calendar fields for WindFM while keeping the synthetic timestamp path independent of the original index.
clipfloat, default=5.0: Input normalization clipping value passed to WindFMPredictor.
predict_kwargsdict or None, default=None: Additional keyword arguments passed directly to WindFMPredictor.predict. Examples are T, top_k, top_p, sample_count, and verbose.
deterministicbool, default=False: Whether predictions should reset the PyTorch random seed before autoregressive sampling.

Attributes:

cutoff: Cut-off = “present time” state of forecaster.
fh: Forecasting horizon that was passed.
is_fitted: Whether fit has been called.
state: State of the estimator.

Notes

WindFMForecaster expects y to contain the target wind power series and X to contain the historical weather covariates required by WindFM: wind speed, wind direction, density, temperature, and pressure. WindFM generates sample paths internally; this wrapper returns the median across those samples as a point forecast.

WindFM was trained with UTC timestamps. If the input index is not datetime-like, synthetic timestamps are generated from start and freq.

References

[1]

WindFM GitHub repository: https://github.com/shiyu-coder/WindFM

[2]

Hang Fan, Yu Shi, Zongliang Fu, Shuo Chen, Wei Wei, Wei Xu, Jian Li (2025). WindFM: An Open-Source Foundation Model for Zero-Shot Wind Power Forecasting. arXiv. https://arxiv.org/abs/2509.06311

[3]

WindFM model card: https://huggingface.co/NeoQuasar/WindFM

[4]

WindFM-robust model card: https://huggingface.co/NeoQuasar/WindFM-robust

[5]

WindFM-Tokenizer model card: https://huggingface.co/NeoQuasar/WindFM-Tokenizer

[6]

WindFM-Tokenizer-robust model card: https://huggingface.co/NeoQuasar/WindFM-Tokenizer-robust

Examples

>>> import pandas as pd
>>> import torch
>>> from sktime.forecasting.base import ForecastingHorizon
>>> from sktime.forecasting.windfm import WindFMForecaster
>>> df = pd.read_csv(
...     "https://raw.githubusercontent.com/shiyu-coder/WindFM/"
...     "refs/heads/master/examples/data/121522.csv",
...     parse_dates=["time"],
...     index_col="time",
... )
>>> lookback, pred_len = 240, 80
>>> covariate_cols = [
...     "wind_speed",
...     "wind_direction",
...     "density",
...     "temperature",
...     "pressure",
... ]
>>> y_train = df["power"].iloc[:lookback]
>>> X_train = df[covariate_cols].iloc[:lookback]
>>> fh = ForecastingHorizon(
...     df.index[lookback : lookback + pred_len],
...     is_relative=False,
... )
>>> device = "cuda:0" if torch.cuda.is_available() else "cpu"
>>> forecaster = WindFMForecaster(
...     model_path="NeoQuasar/WindFM",
...     tokenizer_path="NeoQuasar/WindFM-Tokenizer",
...     # robust variants:
...     # model_path="NeoQuasar/WindFM-robust",
...     # tokenizer_path="NeoQuasar/WindFM-Tokenizer-robust",
...     device=device,
...     deterministic=True,
...     predict_kwargs={
...         "T": 1.0,
...         "top_k": 0,
...         "top_p": 0.9,
...         "sample_count": 100,
...         "verbose": True,
...     },
... )
>>> forecaster.fit(y_train, X=X_train)
WindFMForecaster(...)
>>> y_pred = forecaster.predict(fh=fh)
>>> y_quantiles = forecaster.predict_quantiles(
...     fh=fh,
...     alpha=[0.1, 0.5, 0.9],
... )

Methods

`check_is_fitted`([method_name])	Check if the estimator has been fitted.
`clone`()	Obtain a clone of the object with same hyper-parameters and config.
`clone_tags`(estimator[, tag_names])	Clone tags from another object as dynamic override.
`create_test_instance`([parameter_set])	Construct an instance of the class, using first test parameter set.
`create_test_instances_and_names`([parameter_set])	Create list of all test instances and a list of names for them.
`fit`(y[, X, fh])	Fit forecaster to training data.
`fit_predict`(y[, X, fh, X_pred])	Fit and forecast time series at future horizon.
`get_class_tag`(tag_name[, tag_value_default])	Get class tag value from class, with tag level inheritance from parents.
`get_class_tags`()	Get class tags from class, with tag level inheritance from parent classes.
`get_config`()	Get config flags for self.
`get_fitted_params`([deep])	Get fitted parameters.
`get_param_defaults`()	Get object's parameter defaults.
`get_param_names`([sort])	Get object's parameter names.
`get_params`([deep])	Get a dict of parameters values for this object.
`get_pretrained_params`([deep])	Get pretrained parameters of this estimator.
`get_tag`(tag_name[, tag_value_default, ...])	Get tag value from instance, with tag level inheritance and overrides.
`get_tags`()	Get tags from instance, with tag level inheritance and overrides.
`get_test_params`([parameter_set])	Return testing parameter settings for the estimator.
`is_composite`()	Check if the object is composed of other BaseObjects.
`load_from_path`(serial)	Load object from file location.
`load_from_serial`(serial)	Load object from serialized memory container.
`predict`([fh, X])	Forecast time series at future horizon.
`predict_interval`([fh, X, coverage])	Compute/return prediction interval forecasts.
`predict_proba`([fh, X, marginal])	Compute/return fully probabilistic forecasts.
`predict_quantiles`([fh, X, alpha])	Compute/return quantile forecasts.
`predict_residuals`([y, X])	Return residuals of time series forecasts.
`predict_var`([fh, X, cov])	Compute/return variance forecasts.
`pretrain`(y[, X, fh])	Pre-train forecaster on panel (global) data.
`reset`()	Reset the object to a clean post-init state.
`save`([path, serialization_format])	Save serialized self to bytes-like object or to (.zip) file.
`score`(y[, X, fh])	Scores forecast against ground truth, using MAPE (non-symmetric).
`set_config`(**config_dict)	Set config flags to given values.
`set_params`(**params)	Set the parameters of this object.
`set_random_state`([random_state, deep, ...])	Set random_state pseudo-random seed parameters for self.
`set_tags`(**tag_dict)	Set instance level tag overrides to given values.
`update`(y[, X, update_params])	Update cutoff value and, optionally, fitted parameters.
`update_predict`(y[, cv, X, update_params, ...])	Make predictions and update model iteratively over the test set.
`update_predict_single`([y, fh, X, update_params])	Update model with new data and make forecasts.