Some facts about the PM-M Prototype Selection System
Did you know that the PM-M System can complete a 1 X 10-fold cross-validation test which is conducted on a half a million sample dataset on a regular notebook in about one minute?
Did you know that it takes only about 5 seconds to build (i.e. train) a 10-IB1 based PM-M model ensemble on a half a million sample dataset on a regular notebook?
Did you know that it can take only about five cost function calls for the PM-M system to converge to a nearly zero root mean squared training error? If you think that this is impossible look here for the description and more information about the EkPMinimizer optimization engine!
Moreover, the PM-M system can use an arbitrary underlying classifier to select prototypes unlike most other methods which rely on the k-Nearest Neighbor rule.
The size of a target prototype set obtained with running the PM-M system can be controlled in advance (i.e. one may decide how many instances to retain).