>>This seems to be very nearly the same as the effects of a best-fit >>2-state PFSMs. > >I guess F[inite] S[tate] M[achine], but P? P[artitioning]? "Probabilistic". A basic FSM simply says whether or not a symbol is legal in a state. A PFSM also gives the relative frequency with which legal symbols occur. The expected frequencies are necessary to compute information rates. >>The PFSM partitions the characters into two groups so as >>to optimize the ability to predict the next character based solely on >>which group contains the previous character. > >Another way of achieving the same aim, isn't it? Seems to be, empirically.
<<winmail.dat>>