bagging

EvolutionaryBaggingRegressor(model=AUTOML_REGRESSION_PIPELINE, param_grid=REGRESSION_PARAM_GRID, population_size=10, sampling_size=1, metric=metrics.MSE, sampling_rate=1000, seed=42)

Bases: EvolutionaryBaggingEstimator, base.Regressor

Evolutionary Bagging Regressor follows the Oza Bagging approach to update the population of estimator pipelines.

PARAMETER	DESCRIPTION
model	A river model or model pipeline that can be configured by the parameter grid. DEFAULT: AUTOML_REGRESSION_PIPELINE
param_grid	A parameter grid, that represents the configuration space of the model. DEFAULT: REGRESSION_PARAM_GRID
population_size	The population size estimates the size of the population as well as the size of the ensemble used for the prediction. DEFAULT: 10
sampling_size	The sampling size estimates how many models are mutated within one mutation step. DEFAULT: 1
metric	The river metric that should be optimised. DEFAULT: metrics.MSE
sampling_rate	The sampling rate estimates the number of samples that are executed before a mutation step takes place. DEFAULT: 1000
seed	Random number generator seed for reproducibility. DEFAULT: 42

Examples:

>>> from river import datasets, ensemble, evaluate, metrics, optim, compose
>>> from river import preprocessing, tree, neighbors
>>> from EvOAutoML import regression, pipelinehelper
>>> dataset = datasets.TrumpApproval()
>>> model = regression.EvolutionaryBaggingRegressor(
...     model=compose.Pipeline(
...         ('Scaler', pipelinehelper.PipelineHelperTransformer([
...             ('StandardScaler', preprocessing.StandardScaler()),
...             ('MinMaxScaler', preprocessing.MinMaxScaler()),
...             ('MinAbsScaler', preprocessing.MaxAbsScaler()),
...         ])),
...         ('Regressor', pipelinehelper.PipelineHelperRegressor([
...             ('HT', tree.HoeffdingTreeRegressor()),
...             ('KNN', neighbors.KNNRegressor()),
...         ]))
...         ),
...     param_grid={
...     'Regressor': AUTOML_REGRESSION_PIPELINE
...         .steps['Regressor'].generate({
...             'HT__binary_split': [True, False],
...             'HT__max_depth': [10, 30, 60, 10, 30, 60],
...             'HT__grace_period': [10, 100, 200, 10, 100, 200],
...             'HT__max_size': [5, 10],
...             'KNN__n_neighbors': [1, 5, 20],
...             'KNN__window_size': [100, 500, 1000],
...             'KNN__p': [1, 2]
...         })
... },
... seed=42
... )
>>> metric = metrics.MSE()
>>> for x, y in dataset:
...     y_pred = model.predict_one(x)  # make a prediction
...     metric = metric.update(y, y_pred)  # update the metric
...     model = model.learn_one(x,y)  # make the model learn
>>> print(f'MSE: {metric.get():.2f}')
MSE: 2.35

predict_one(x)

Averages the predictions of each regressor.