In Gaussian discrimination, it is crucial to define handled
data in terms of two samples: a learning sample L and a
prediction sample P, resulting respectively from the following
subpopulations: Ω and Ω∗. In our case these two subpopulations
are different.