openstef.model package

Subpackages

Submodules

openstef.model.basecase module

class openstef.model.basecase.BaseCaseModel

Bases: BaseEstimator, RegressorMixin

property can_predict_quantiles
fit()
static make_basecase_forecast(forecast_input_data, overwrite_delay_hours=48)

Make a basecase forecast.

The idea of the basecase forecast is that if all else fails, this forecasts is still available. Basecase example: the load of last week.

Parameters:

  • forecast_input_data (DataFrame) – Forecast input dataframe

  • overwrite_delay_hours (int) – times before this in the future are not forecasted

Raises:

  • ValueError – if columns T-7d or T-14d is not present

  • ValueError – If the start of the forecast is before the horizon of the regular forecast

Return type:

DataFrame

Returns:

Basecase forecast

predict(forecast_input_data)

Predict using the basecase method. The basecase forecast is determined by the T-7d and T-14d load.

This means fitting the model is not required. However a fit method is still included to be fully comatible with sklearn.

Parameters:

forecast_input_data (DataFrame) – Forecast input dataframe

Return type:

DataFrame

Returns:

Basecase forecast

set_predict_request(*, forecast_input_data: bool | None | str = '$UNCHANGED$') → BaseCaseModel

Request metadata passed to the predict method.

Note that this method is only relevant if enable_metadata_routing=True (see sklearn.set_config()). Please see User Guide on how the routing mechanism works.

The options for each parameter are:

  • True: metadata is requested, and passed to predict if provided. The request is ignored if metadata is not provided.

  • False: metadata is not requested and the meta-estimator will not pass it to predict.

  • None: metadata is not requested, and the meta-estimator will raise an error if the user provides it.

  • str: metadata should be passed to the meta-estimator with this given alias instead of the original name.

The default (sklearn.utils.metadata_routing.UNCHANGED) retains the existing request. This allows you to change the request for some parameters and not others.

Added in version 1.3.

Note

This method is only relevant if this estimator is used as a sub-estimator of a meta-estimator, e.g. used inside a Pipeline. Otherwise it has no effect.

Parameters:

forecast_input_data (str, True, False, or None, default=sklearn.utils.metadata_routing.UNCHANGED) – Metadata routing for forecast_input_data parameter in predict.

Returns:

self – The updated object.

Return type:

object

set_score_request(*, sample_weight: bool | None | str = '$UNCHANGED$') → BaseCaseModel

Request metadata passed to the score method.

Note that this method is only relevant if enable_metadata_routing=True (see sklearn.set_config()). Please see User Guide on how the routing mechanism works.

The options for each parameter are:

  • True: metadata is requested, and passed to score if provided. The request is ignored if metadata is not provided.

  • False: metadata is not requested and the meta-estimator will not pass it to score.

  • None: metadata is not requested, and the meta-estimator will raise an error if the user provides it.

  • str: metadata should be passed to the meta-estimator with this given alias instead of the original name.

The default (sklearn.utils.metadata_routing.UNCHANGED) retains the existing request. This allows you to change the request for some parameters and not others.

Added in version 1.3.

Note

This method is only relevant if this estimator is used as a sub-estimator of a meta-estimator, e.g. used inside a Pipeline. Otherwise it has no effect.

Parameters:

sample_weight (str, True, False, or None, default=sklearn.utils.metadata_routing.UNCHANGED) – Metadata routing for sample_weight parameter in score.

Returns:

self – The updated object.

Return type:

object

openstef.model.confidence_interval_applicator module

class openstef.model.confidence_interval_applicator.ConfidenceIntervalApplicator(model, forecast_input_data)

Bases: object

add_confidence_interval(forecast, pj)

Add a confidence interval to a forecast.

Adds a confidence interval to a forecast in two ways:
  1. “stdev” column, this is a column with a standard deviation that is

    determined during training (ConfidenceGenerator)

  2. Quantile columns, these columns give a more precise defenition of the

    confidence interval. Quantile columns are determined with one of two methods, depending on the model type group:

    1. Default, using the “stdev” column and the assumption the error is

      normally distributed.

    2. Quantile regression, this method is only available for quantile

      models and uses specifically trained models to estimate the quantiles of the confidence interval.

    Depending on the model type (quantile or non quantile),

    a confidence interval is added to the forecast based on quantile regression or the default method.

Parameters:

  • forecast (DataFrame) – Forecast DataFrame with columns: “forecast”

  • pj (PredictionJobDataClass) – Prediction job

Return type:

DataFrame

Returns:

Forecast DataFrame with columns; “forecast”, “stdev” and quantile columns.

openstef.model.fallback module

openstef.model.fallback.generate_fallback(forecast_input, load, fallback_strategy='extreme_day')

Make a fall back forecast, Set the value of the forecast ‘quality’ column to ‘substituted’.

Currently only fallback_strategy=extreme day is implemented which return historic profile of most extreme day.

Parameters:

  • forecast_input (DataFrame) – dataframe desired for the forecast

  • load (DataFrame) – index=datetime, columns=[‘load’]

  • fallback_strategy (str) – strategy to determine fallback. options: - extreme_day: use daily profile of most extreme day

Return type:

DataFrame

Returns:

Fallback forecast DataFrame with columns; ‘forecast’, ‘quality’

Raises:

  • ValueError if len(load) == 0 –

  • NotImplementedError if fallback_strategy != 'extreme_day' –

openstef.model.model_creator module

class openstef.model.model_creator.ModelCreator

Bases: object

Factory object for creating machine learning models.

MODEL_CONSTRUCTORS = {MLModelType.ARIMA: <class 'openstef.model.regressors.arima.ARIMAOpenstfRegressor'>, MLModelType.LGB: <class 'openstef.model.regressors.lgbm.LGBMOpenstfRegressor'>, MLModelType.LINEAR: <class 'openstef.model.regressors.linear.LinearOpenstfRegressor'>, MLModelType.LINEAR_QUANTILE: <class 'openstef.model.regressors.linear_quantile.LinearQuantileOpenstfRegressor'>, MLModelType.XGB: <class 'openstef.model.regressors.xgb.XGBOpenstfRegressor'>, MLModelType.XGB_MULTIOUTPUT_QUANTILE: <class 'openstef.model.regressors.xgb_multioutput_quantile.XGBMultiOutputQuantileOpenstfRegressor'>, MLModelType.XGB_QUANTILE: <class 'openstef.model.regressors.xgb_quantile.XGBQuantileOpenstfRegressor'>}
static create_model(model_type, **kwargs)

Create a machine learning model based on model type.

Parameters:

  • model_type (Union[MLModelType, str]) – Model type to construct.

  • kwargs – Optional keyword argument to pass to the model.

Raises:

NotImplementedError – When using an invalid model_type.

Return type:

OpenstfRegressor

Returns:

OpenSTEF model

openstef.model.objective module

class openstef.model.objective.ARIMARegressorObjective(*args, **kwargs)

Bases: RegressorObjective

get_params(trial)

Get parameters for ARIMA Regressor Objective with objective specific parameters.

Temporary, it seems strange to use optuna for ARIMA models, it is usually done via statistical analysis and heuristics.

Args: trial

Return type:

dict

Returns:

Dictionary with hyperparameter name as key and hyperparamer value as value.

class openstef.model.objective.LGBRegressorObjective(*args, **kwargs)

Bases: RegressorObjective

get_params(trial)

Get parameters for LGB Regressor Objective with objective specific parameters.

Args: trial

Return type:

dict

Returns:

Dictionary with hyperparameter name as key and hyperparamer value as value.

get_pruning_callback(trial)
class openstef.model.objective.LinearRegressorObjective(*args, **kwargs)

Bases: RegressorObjective

get_params(trial)

Get parameters for Linear Regressor Objective with objective specific parameters.

Args: trial

Return type:

dict

Returns:

Dictionary with hyperparameter name as key and hyperparamer value as value.

class openstef.model.objective.RegressorObjective(model, input_data, split_func=None, split_args=None, test_fraction=0.15, validation_fraction=0.15, eval_metric='mae', verbose=False)

Bases: object

Regressor optuna objective function.

Use any of the derived classes for optimization using an optuna study. The constructor is used to set the “input_data”, specify the splitting function and its arguments and optionally add some configuration. Next the instance will be called by he optuna study during optimization.

Example usage:

# initialize a (derived class) objective function
objective = XGBRegressorObjective(input_data, test_fraction)
# use the objective function
study.optimize(objective)
create_report(model)

Generate a report from the data available inside the objective function.

Parameters:

model (OpenstfRegressor) – OpenstfRegressor, model to create a report on

Return type:

Report

Returns:

Report about the model

classmethod get_default_values()
Return type:

dict

get_params(trial)

Get parameters for objective without model specific get_params function.

Args: trial

Return type:

dict

Returns:

Dictionary with hyperparameter name as key and hyperparamer value as value.

get_pruning_callback(trial)
get_trial_track()

Get a dictionary of al trials.

Return type:

dict

Returns:

Dict with al trials and it’s parameters

class openstef.model.objective.XGBMultioutputQuantileRegressorObjective(*args, **kwargs)

Bases: RegressorObjective

get_params(trial)

Get parameters for XGB Multioutput Quantile Regressor Objective with objective specific parameters.

Args: trial

Return type:

dict

Returns:

Dictionary with hyperparameter name as key and hyperparamer value as value.

get_pruning_callback(trial)
class openstef.model.objective.XGBQuantileRegressorObjective(*args, **kwargs)

Bases: RegressorObjective

get_params(trial)

Get parameters for XGBQuantile Regressor Objective with objective specific parameters.

Args: trial

Return type:

dict

Returns:

Dictionary with hyperparameter name as key and hyperparamer value as value.

get_pruning_callback(trial)
class openstef.model.objective.XGBRegressorObjective(*args, **kwargs)

Bases: RegressorObjective

classmethod get_default_values()
Return type:

dict

get_params(trial)

Get parameters for XGB Regressor Objective with objective specific parameters.

Args: trial

Return type:

dict

Returns:

Dictionary with hyperparameter name as key and hyperparamer value as value.

get_pruning_callback(trial)

openstef.model.objective_creator module

class openstef.model.objective_creator.ObjectiveCreator

Bases: object

OBJECTIVES = {MLModelType.ARIMA: <class 'openstef.model.objective.ARIMARegressorObjective'>, MLModelType.LGB: <class 'openstef.model.objective.LGBRegressorObjective'>, MLModelType.LINEAR: <class 'openstef.model.objective.LinearRegressorObjective'>, MLModelType.LINEAR_QUANTILE: <class 'openstef.model.objective.LinearRegressorObjective'>, MLModelType.XGB: <class 'openstef.model.objective.XGBRegressorObjective'>, MLModelType.XGB_MULTIOUTPUT_QUANTILE: <class 'openstef.model.objective.XGBMultioutputQuantileRegressorObjective'>, MLModelType.XGB_QUANTILE: <class 'openstef.model.objective.XGBQuantileRegressorObjective'>}
static create_objective(model_type)

Create an objective function based on model type.

Parameters:

model_type (Union[MLModelType, str]) – Model type to construct.

Raises:

NotImplementedError – When using an invalid model_type.

Return type:

RegressorObjective

Returns:

Objective function

openstef.model.serializer module

class openstef.model.serializer.MLflowSerializer(mlflow_tracking_uri)

Bases: object

get_model_age(experiment_name, hyperparameter_optimization_only=False)

Get model age of most recent model.

Parameters:

  • experiment_name (str) – Name of the experiment, often the id of the predition job.

  • hyperparameter_optimization_only (bool) – Set to true if only hyperparameters optimaisation events should be considered.

Return type:

int

load_model(experiment_name)

Load sklearn compatible model from MLFlow.

Parameters:

experiment_name (str) – Name of the experiment, often the id of the predition job.

Raises:

LookupError – If model is not found in MLflow.

Return type:

tuple[OpenstfRegressor, ModelSpecificationDataClass]

remove_old_models(experiment_name, max_n_models=10)

Remove old models per experiment.

save_model(model, experiment_name, model_type, model_specs, report, phase='training', **kwargs)

Save sklearn compatible model to MLFlow.

Return type:

None

openstef.model.standard_deviation_generator module

class openstef.model.standard_deviation_generator.StandardDeviationGenerator(validation_data)

Bases: object

generate_standard_deviation_data(model)

Generate the standard data.

Calculates the difference between realised and predicted on validation set. For each hour of the day the std of the difference is calculated.

Parameters:

model (RegressorMixin) – The trained model

Return type:

RegressorMixin

Returns:

The model with the std data added.

Module contents