Interrating agreement
A simple extension of the measures discussed in the previous section on the ensemble classifiers is to compute the measures for all possible pairs of the ensemble and then simply average over all those values. This task constitutes the next exercise.
Exercise: For all possible combinations of ensemble pairs, calculate the disagreement measure, Yule's statistic, correlation coefficient, Cohen's kappa, and the double-fault measure. After doing this, obtain the average of the comparisons and report them as the ensemble diversity.
Here, we will propose alternative measures of diversity and kick-start the discussion with the entropy measure. In all discussions in this section, we will use the oracle outputs.
Entropy measure
You may recall that we denote the oracle outputs according to 
. For a particular instance, the ensemble is most diverse if the number of classifiers misclassifying it is 
. This means that 
of the 
s are 0s, and the rest of the 
, 
s are 1s. The entropy...
