Possibly numerical stable (even on recursive) generic predictor on discrete Lebesgue measurable expectation value. Discrete Lebesgue measurable condition needs deterministic input on the range invariant defined (invariant might be categorized including 0 vector). Normally, if we calculate first order prediction, the invariant range we suppose into argument needs to be larger than inner status dimension the function have. Applying recursive this predictor causes increase status dimension to be calculated and compete with jammers, but statistical illegal result will increase.
Every predictor has its jammer, instead of one predictor, this outputs 2 of choices. If we need longer span differed to status samples use with predictor, we need 3 of choices.
This predictor uses only first order categorization. If we need second order or more with statistical switching, it is recommended to use p1, p2 but they need to specify the status length to be used explicitly.