Comparing dart

The base learner dart is similar to gbtree in the sense that both are gradient boosted trees. The primary difference is that dart removes trees (called dropout) during each round of boosting.

In this section, we will apply and compare the base learner dart to other base learners in regression and classification problems.

DART with XGBRegressor

Let's see how dart performs on the Diabetes dataset:

1. First, redefine X and y using load_diabetes as before:

   X, y = load_diabetes(return_X_y=True)

2. To use dart as the XGBoost base learner, set the XGBRegressor parameter booster='dart' inside the regression_model function:

   regression_model(XGBRegressor(booster='dart', objective='reg:squarederror'))

   The score is as follows:

   65.96444746130739

The dart base learner gives the same result as the gbtree base learner down to two decimal places. The similarity of results is on account of the small dataset and the success of the gbtree...