This variant is a straightforward
combination of GM-CDW and EF-CDW. The instances are transformed to the
expanded-features format, the standard CDW algorithm is applied, and the
expanded-feature weights on the classes are averaged to get global weights for
each expanded feature. This variant exhibits test case locality but not class locality.
It should do especially well on tasks where only certain feature values are
relevant, but the relevant values do not vary from class to class