The five dimensions constitute 5 informative features.
15 linear combinations of those features are added to form a set of 20
(redundant) informative features.
There are 480 distractor features called 'probes' having no predictive
power.Included is the original training set with label -1 changed to 0.