Class measuring change in EMP under AIC from splitting 2 nodes. More...
Class measuring change in EMP under AIC from splitting 2 nodes.
Under AIC, EMP is -1 x sum over leaves of (counts in leaf x (ln(count in leaf /(n x vol of leaf))) where n is the total number of data points in the histogram
For two leaf nodes we are comparing the change in -1 x the sum over leaves of (counts in leaf x (ln(count in leaf /(n x vol of leaf) which would result if each node were to be the one to be split.
The smaller (more negative) the value returned by getSplitChangeEMPAIC(), the more a node will reduce or least increase the overall EMP by being split, so it should be higher, ie more to right, in the ordering, so we meausure using the negated value.