Hello ajb, Thursday, July 12, 2007, 12:16:22 PM, you wrote:
well, ppm differs from simple order-0 coder in that it uses previous symbols to calculate probability. lzw differs from order-0 coder with flat static probabilities in that it encodes a whole word each time.
oh, i just recalled that flat distribution conventionally called order -1 model. so starting from non-compressible order -1 model, lzw and ppm goes into opposite direction: * ppm uses several previous chars to predict one next character * lz77/78 predicts (in the order -1 model) several next chars at once if you interested, there are many algorithms combining both approaches: lzrw5/rolz, lzp, lzcb, lzma -- Best regards, Bulat mailto:Bulat.Ziganshin@gmail.com