ForecastingBenchmark

class ForecastingBenchmark(id_format: str | None = None, backend=None, backend_params=None, return_data=False)

Forecasting benchmark.

Run a series of forecasters against a series of tasks defined via dataset loaders, cross validation splitting strategies and performance metrics, and return results as a df (as well as saving to file).

Parameters:

id_format: str, optional (default=None)

A regex used to enforce task/estimator ID to match a certain format

backendstring, by default “None”.

Parallelization backend to use for runs.

• “None”: executes loop sequentially, simple list comprehension

• “loky”, “multiprocessing” and “threading”: uses joblib.Parallel loops

• “joblib”: custom and 3rd party joblib backends, e.g., spark

• “dask”: uses dask, requires dask package in environment

• “dask_lazy”: same as “dask”, but changes the return to (lazy)

  dask.dataframe.DataFrame.

• “ray”: uses ray, requires ray package in environment

Recommendation: Use “dask” or “loky” for parallel evaluate. “threading” is unlikely to see speed ups due to the GIL and the serialization backend (cloudpickle) for “dask” and “loky” is generally more robust than the standard pickle library used in “multiprocessing”.

backend_paramsdict, optional

additional parameters passed to the backend as config. Directly passed to utils.parallel.parallelize. Valid keys depend on the value of backend:

• “None”: no additional parameters, backend_params is ignored

• “loky”, “multiprocessing” and “threading”: default joblib backends

any valid keys for joblib.Parallel can be passed here, e.g., n_jobs, with the exception of backend which is directly controlled by backend. If n_jobs is not passed, it will default to -1, other parameters will default to joblib defaults. - “joblib”: custom and 3rd party joblib backends, e.g., spark. any valid keys for joblib.Parallel can be passed here, e.g., n_jobs, backend must be passed as a key of backend_params in this case. If n_jobs is not passed, it will default to -1, other parameters will default to joblib defaults. - “dask”: any valid keys for dask.compute can be passed, e.g., scheduler

• “ray”: The following keys can be passed:

  • “ray_remote_args”: dictionary of valid keys for ray.init

  • “shutdown_ray”: bool, default=True; False prevents ray from shutting

    down after parallelization.

  • “logger_name”: str, default=”ray”; name of the logger to use.

  • “mute_warnings”: bool, default=False; if True, suppresses warnings

return_databool, optional (default=False)

Whether to return the prediction and the ground truth data in the results.

Methods

add_estimator(estimator[, estimator_id])	Register an estimator to the benchmark.
add_task(dataset_loader, cv_splitter, scorers)	Register a forecasting task to the benchmark.
run([output_file, force_rerun])	Run the benchmarking for all tasks and estimators.

add_task(dataset_loader: Callable | tuple, cv_splitter: BaseSplitter, scorers: list[BaseMetric], task_id: str | None = None, cv_global: BaseSplitter | None = None, error_score: str = 'raise', strategy: str = 'refit', cv_global_temporal: SingleWindowSplitter | None = None)

Register a forecasting task to the benchmark.

Parameters:

dataUnion[Callable, tuple]

Can be

• a function which returns a dataset, like from sktime.datasets.

• a tuple containing two data container that are sktime comptaible.

• single data container that is sktime compatible (only endogenous data).

cv_splitterBaseSplitter object

Splitter used for generating validation folds.

scorersa list of BaseMetric objects

Each BaseMetric output will be included in the results.

task_idstr, optional (default=None)

Identifier for the benchmark task. If none given then uses dataset loader name combined with cv_splitter class name.

cv_global: sklearn splitter, or sktime instance splitter, default=None

If cv_global is passed, then global benchmarking is applied, as follows:

1. the cv_global splitter is used to split data at instance level, into a global training set y_train, and a global test set y_test_global. 2. The estimator is fitted to the global training set y_train. 3. cv_splitter then splits the global test set y_test_global temporally, to obtain temporal splits y_past, y_true.

Overall, with y_train, y_past, y_true as above, the following evaluation will be applied:

forecaster.fit(y=y_train, fh=cv.fh)
y_pred = forecaster.predict(y=y_past)
metric(y_true, y_pred)

error_score“raise” or numeric, default=np.nan

Value to assign to the score if an exception occurs in estimator fitting. If set to “raise”, the exception is raised. If a numeric value is given, FitFailedWarning is raised.

strategy{“refit”, “update”, “no-update_params”}, optional, default=”refit”

defines the ingestion mode when the forecaster is updated with new data

• “refit” = forecaster is refitted to each training window

• “update” = forecaster is updated with training window data, in sequence provided

• “no-update_params” = fit to first training window, re-used without fit or update

cv_global_temporal: SingleWindowSplitter, default=None

ignored if cv_global is None. If passed, it splits the Panel temporally before the instance split from cv_global is applied. This avoids temporal leakage in the global evaluation across time series. Has to be a SingleWindowSplitter. cv is applied on the test set of the combined application of cv_global and cv_global_temporal.

Returns:

A dictionary of benchmark results for that forecaster

add_estimator(estimator: BaseEstimator, estimator_id: str | None = None)

Register an estimator to the benchmark.

Parameters:

estimatordict, list or BaseEstimator object

Estimator to add to the benchmark.

• if BaseEstimator, single estimator. estimator_id is generated as the estimator’s class name if not provided.

• If dict, keys are ``estimator_id``s used to customise identifier ID and values are estimators.

• If list, each element is an estimator. ``estimator_id``s are generated automatically using the estimator’s class name.

estimator_idstr, optional (default=None)

Identifier for estimator. If none given then uses estimator’s class name.

run(output_file: str = None, force_rerun: str | list[str] = 'none')

Run the benchmarking for all tasks and estimators.

If output_file is provided, results will be saved to a file or location, in a format inferred from the file extension.

The exact format is determined by the storage backend used, see documentation on storage handlers in sktime.benchmarking._storage_handlers.get_storage_backend.

Parameters:

output_filestr or None (default)

Path to save the results to. If None, results will not be saved.

force_rerunUnion[str, list[str]], optional (default=”none”)

• If “none”, will skip validation if results already exist.

• If “all”, will run validation for all tasks and models.

• If list of str, will run validation for tasks and models in list.