ash

ASH is basically a rewrite of PPMY

In ash there're more compact tree [compared to PPMY], better weighting function (with automatically tuned parameters), unary coding with SSE for specified number of high-ranked symbols (instead of always mixing whole distributions) - and, now, E8 and better memory overflow handling.

[ASH and PPMY] only have some chances with texts, as there's no sparse contexts or structure handling, and simple bytewise models are generally inferior to bitwise on nonstationary data, as bitwise models adapt faster (8 updates per byte instead of 1).

The 3 configurations were submitted by the author in April 5, 2009.

(Have info that should be added here? E-mail.)