ForecastingWorkflowConfig#
- class openstef_models.presets.ForecastingWorkflowConfig(*, kind: ~typing.Literal['single'] = 'single', model_id: ~openstef_models.mixins.model_serializer.ModelIdentifier, run_name: str | None = None, model: ~typing.Literal['xgboost', 'gblinear', 'flatliner', 'median', 'constant_quantile', 'lgbm', 'lgbmlinear'], quantiles: list[~openstef_core.types.Quantile] = [0.5], sample_interval: ~datetime.timedelta = datetime.timedelta(seconds=900), horizons: list[~openstef_core.types.LeadTime] = [LeadTime('P2D')], xgboost_hyperparams: ~openstef_models.models.forecasting.xgboost_forecaster.XGBoostHyperParams = XGBoostHyperParams(n_estimators=100, learning_rate=0.3, max_depth=6, min_child_weight=1, gamma=0, objective='pinball_loss', evaluation_metric='mean_pinball_loss', reg_alpha=0, reg_lambda=1, max_delta_step=0, max_leaves=0, grow_policy='depthwise', max_bin=256, num_parallel_trees=1, subsample=1.0, colsample_bytree=1.0, colsample_bylevel=1.0, colsample_bynode=1.0, tree_method='auto', random_state=42, early_stopping_rounds=None, use_target_scaling=True), gblinear_hyperparams: ~openstef_models.models.forecasting.gblinear_forecaster.GBLinearHyperParams = GBLinearHyperParams(n_steps=500, updater='shotgun', learning_rate=0.15, objective='reg:quantileerror', evaluation_metric='mean_pinball_loss', reg_alpha=0.0001, reg_lambda=0.1, feature_selector='shuffle', top_k=0, random_state=None, early_stopping_rounds=10), lgbm_hyperparams: ~openstef_models.models.forecasting.lgbm_forecaster.LGBMHyperParams = LGBMHyperParams(n_estimators=100, learning_rate=0.49, max_depth=2, min_child_weight=1, min_data_in_leaf=10, min_data_in_bin=10, reg_alpha=0, reg_lambda=1, num_leaves=100, max_bin=256, colsample_bytree=1.0, early_stopping_rounds=None), lgbmlinear_hyperparams: ~openstef_models.models.forecasting.lgbmlinear_forecaster.LGBMLinearHyperParams = LGBMLinearHyperParams(n_estimators=100, learning_rate=0.07, max_depth=6, min_child_weight=0.06, min_data_in_leaf=500, min_data_in_bin=500, reg_alpha=0, reg_lambda=1, num_leaves=30, max_bin=256, colsample_bytree=1, early_stopping_rounds=None), location: ~openstef_models.presets.forecasting_workflow.LocationConfig = LocationConfig(name='test_location', description='', coordinate=Coordinate(latitude=52.132633, longitude=5.291266), country_code='NL'), school_holidays: ~openstef_models.transforms.time_domain.school_holiday_features_adder.SchoolHolidayConfig | None = None, target_column: str = 'load', energy_price_column: str = 'day_ahead_electricity_price', radiation_column: str = 'radiation', wind_speed_column: str = 'windspeed', pressure_column: str = 'pressure', temperature_column: str = 'temperature', relative_humidity_column: str = 'relative_humidity', selected_features: ~openstef_models.utils.feature_selection.FeatureSelection = FeatureSelection(include=None, include_regex=None, exclude=None, exclude_regex=None), predict_history: ~datetime.timedelta = datetime.timedelta(days=14), cutoff_history: ~datetime.timedelta = datetime.timedelta(0), completeness_threshold: float = 0.5, completeness_threshold_target_constant_quantile: float = 0.03, flatliner_threshold: ~datetime.timedelta = datetime.timedelta(days=1), detect_non_zero_flatliner: bool = False, predict_nonzero_flatliner: bool = False, shifters: list[~openstef_models.transforms.general.shifter.Shifter] = [], rolling_aggregate_features: list[~openstef_models.transforms.time_domain.rolling_aggregates_adder.AggregationFunction] = [], clip_features: ~openstef_models.utils.feature_selection.FeatureSelection = FeatureSelection(include=None, include_regex=None, exclude=None, exclude_regex=None), nan_on_outlier_features: ~openstef_models.utils.feature_selection.FeatureSelection = FeatureSelection(include=set(), include_regex=set(), exclude=None, exclude_regex=None), outlier_strictness_n_std: float = 2.0, max_day_lags: int = 14, sample_weight_config: ~openstef_models.transforms.general.sample_weighter.SampleWeightConfig = <factory>, data_splitter: ~openstef_models.utils.data_split.DataSplitter = DataSplitter(val_fraction=0.15, test_fraction=0.0, stratification_fraction=0.15, min_days_for_stratification=4, random_state=42), evaluation_metrics: list[~openstef_beam.evaluation.metric_providers.MetricProvider] = <factory>, mlflow_storage: ~openstef_models.integrations.mlflow.mlflow_storage.MLFlowStorage | None = <factory>, model_reuse_enable: bool = True, model_reuse_max_age: ~datetime.timedelta = datetime.timedelta(days=7), model_selection_enable: bool = True, model_selection_metric: tuple[~typing.Union[~openstef_core.types.Quantile, ~typing.Literal['global']], str, ~openstef_beam.evaluation.metric_providers.MetricDirection] = (0.5, 'R2', 'higher_is_better'), model_selection_old_model_penalty: float = 1.2, model_performance_callback_enabled: bool = False, model_performance_callback_metric_threshold: tuple[~typing.Union[~openstef_core.types.Quantile, ~typing.Literal['global']], str, ~openstef_beam.evaluation.metric_providers.MetricDirection, float] = (0.5, 'R2', 'higher_is_better', 0.0), verbosity: ~typing.Literal[0, 1, 2, 3, True] = 0, tags: dict[str, str] = <factory>, experiment_tags: dict[str, str] = <factory>)[source]#
Bases:
BaseConfigConfiguration for forecasting workflows.
Defines all parameters needed to set up a forecasting model, including model type, hyperparameters, location information, data columns, and feature engineering settings.
- Parameters:
kind (
Literal['single'])model_id (
str)model (
Literal['xgboost','gblinear','flatliner','median','constant_quantile','lgbm','lgbmlinear'])sample_interval (
timedelta)xgboost_hyperparams (
XGBoostHyperParams)gblinear_hyperparams (
GBLinearHyperParams)lgbm_hyperparams (
LGBMHyperParams)lgbmlinear_hyperparams (
LGBMLinearHyperParams)location (
LocationConfig)school_holidays (
SchoolHolidayConfig|None)target_column (
str)energy_price_column (
str)radiation_column (
str)wind_speed_column (
str)pressure_column (
str)temperature_column (
str)relative_humidity_column (
str)selected_features (
FeatureSelection)predict_history (
timedelta)cutoff_history (
timedelta)completeness_threshold (
float)completeness_threshold_target_constant_quantile (
float)flatliner_threshold (
timedelta)detect_non_zero_flatliner (
bool)predict_nonzero_flatliner (
bool)shifters (
list[Shifter])rolling_aggregate_features (
list[Literal['mean','median','max','min']])clip_features (
FeatureSelection)nan_on_outlier_features (
FeatureSelection)outlier_strictness_n_std (
float)max_day_lags (
int)sample_weight_config (
SampleWeightConfig)data_splitter (
DataSplitter)evaluation_metrics (
list[MetricProvider])mlflow_storage (
MLFlowStorage|None)model_reuse_enable (
bool)model_reuse_max_age (
timedelta)model_selection_enable (
bool)model_selection_metric (
tuple[Union[Quantile,Literal['global']],str,MetricDirection])model_selection_old_model_penalty (
float)model_performance_callback_enabled (
bool)model_performance_callback_metric_threshold (
tuple[Union[Quantile,Literal['global']],str,MetricDirection,float])
- model_id: ModelIdentifier#
Unique identifier for the forecasting model.
- model: Literal['xgboost', 'gblinear', 'flatliner', 'median', 'constant_quantile', 'lgbm', 'lgbmlinear']#
Type of forecasting model to use.
- xgboost_hyperparams: XGBoostHyperParams#
Hyperparameters for XGBoost forecaster.
- gblinear_hyperparams: GBLinearHyperParams#
Hyperparameters for GBLinear forecaster.
- lgbm_hyperparams: LGBMHyperParams#
Hyperparameters for LightGBM forecaster.
- lgbmlinear_hyperparams: LGBMLinearHyperParams#
Hyperparameters for LightGBM forecaster.
- location: LocationConfig#
Location information for the forecasting workflow.
- selected_features: FeatureSelection#
Feature selection for which features to include/exclude.
- cutoff_history: timedelta#
Amount of historical data to exclude from training and prediction due to incomplete features from lag-based preprocessing. When using lag transforms (e.g., lag-14), the first N days contain NaN values. Set this to match your maximum lag duration (e.g., timedelta(days=14)). Default of 0 assumes no invalid rows are created by preprocessing. Note: should be same as predict_history if you are using lags. We default to disabled to keep the same behaviour as openstef 3.0.
- completeness_threshold: float#
Minimum fraction of data that should be available for making a regular forecast.
- completeness_threshold_target_constant_quantile: float#
Minimum fraction of target data that should be available for making a constant quantile forecast.
- flatliner_threshold: timedelta#
Number of minutes that the load has to be constant to detect a flatliner.
- detect_non_zero_flatliner: bool#
If True, flatliners are also detected on non-zero values (median of the load).
- predict_nonzero_flatliner: bool#
If True, predict the median of load measurements instead of zero (only for flatliner model).
- shifters: list[Shifter]#
List of feature shifts to align aggregation intervals. Each Shifter can target different features with different aggregation periods.
- rolling_aggregate_features: list[AggregationFunction]#
If not None, rolling aggregate(s) of load will be used as features in the model.
- clip_features: FeatureSelection#
Feature selection for which features to clip to their learned range.
- nan_on_outlier_features: FeatureSelection#
Feature selection for which features to replace out-of-range values with NaN. Defaults to no features (disabled).
- outlier_strictness_n_std: float#
Number of standard deviations for outlier detection (lower = stricter bounds). Only used when nan_on_outlier_features is enabled.
- max_day_lags: int#
Maximum number of days to look back for day-based lags. Default is 14 days (two weekly cycles). Set to 7 for a single weekly cycle.
- sample_weight_config: SampleWeightConfig#
Sample weighting configuration. Controls how training samples are weighted. Defaults to weight_exponent=1.0 for gblinear, 0.0 (uniform) for other models.
- data_splitter: DataSplitter#
Configuration for splitting data into training, validation, and test sets.
- mlflow_storage: MLFlowStorage | None#
Configuration for MLflow experiment tracking and model storage.
- model_selection_metric: tuple[Union[Quantile, Literal['global']], str, MetricDirection]#
Metric to monitor for model performance when retraining.
- model_selection_old_model_penalty: float#
Penalty to apply to the old model’s metric to bias selection towards newer models.
- model_performance_callback_enabled: bool#
Whether to enable the ModelPerformanceCallback that evaluates model performance at the end of fitting.