EF-CDW is identical to CDW for domains whose features all have binary
values, e.g., LED-7 and LED-24. This is because feature expansion on binary
features makes two new features with similar distributions. (One is the mirror of
the other.) The two expansion features are each assigned weights that are half
the normal CDW weights, and the relative distances are unchanged. The relative
ranks of unweighted k-NN distances are also unchanged by feature expansion