In tkonopka/mlensemble: Ensembles of Machine Learning Models

Ensembles

Ensembles are collections of ml_model objects that are used together to make predictions.

An ensemble can consist of a single model. Using the m_lm_x1 model defined in the previous section as an example,

e_lm_1 <- ml_ensemble() + m_lm_x1
e_lm_1

This construction first creates an empty ensemble using ml_ensemble() and then adds an existing model into the ensemble. The ensemble can be used with predict as before,

predict(e_lm_1, data_test)

Because this ensemble consists of a single model, the predictions are equivalent to results in the previous section.

However, a key characteristic of ensembles is their support for multiple models. In the present context, it is possible to train a new model on the x2 variable of the synthetic dataset and create an ensemble of two models,

m_lm_x2 <- ml_model(lm(y ~ x2, data=data_train), name="lm_x2")
e_lm_2 <- m_lm_x1 + m_lm_x2
e_lm_2

A similar result can be achieved by extending the existing ensemble with the new model,

e_lm_2 <- e_lm_1 + m_lm_x2
e_lm_2

Predictions using the new ensemble are different,

predict(e_lm_2, data_test)

Importantly, the above command generates a warning. This signals that the ensemble is taking unchecked assumptions during the prediction, which can be addressed by calibration.

tkonopka/mlensemble documentation built on March 19, 2022, 7:28 a.m.