somewhat amused how unreasonably effective PPMd compression actually is.
i used to poke around with that mode in 7zip and compare it to LZMA and they often ended up close together. PPMd would be slightly larger by a megabyte or something on text tests.
PPMd however has an absolutely flat amount of memory for decompression. like 20mb or something. it's just the probability table and a state machine.
LZMA is a hog.
i used to poke around with that mode in 7zip and compare it to LZMA and they often ended up close together. PPMd would be slightly larger by a megabyte or something on text tests.
PPMd however has an absolutely flat amount of memory for decompression. like 20mb or something. it's just the probability table and a state machine.
LZMA is a hog.
Follow
💀
@icedquinn
LZMA is more multi-purpose. When you make something more targeted for one specific thing, it's gonna be more efficient. It's just how things are. 
Never did anything with PPMd myself. Maybe I'll test to see how epubs behave with it compared to LZMA. I wouldn't have any other use for it.
Huffman tables (zip, zlib) are about finding the most common bit patterns and then just recoding those with a shorter one.
PPM is similar to that. It makes a statistical table of the most common bytes which occur in the context of some number of previous bytes. Then it just encodes whether the most common ocurrence in the chain did occur, or the second or third etc.
Text just happens to be an extremely redundant carrier so it compresses well with ... anything.