In the chapter “Applying Decision Trees” page 12/15, some noise is introduced to show an increase in the variance:
income[“noise”] = numpy.random.randint(4, size=income.shape)
However if we use the Classifier parameters from the previous example:
min_samples_split = 13,max_depth = 7
we still get the good results from before (train AUC ~ 0.74 and test AUC ~ 0.77). So I didn’t get the point of this example.