 | Przemyslaw Skibinski integrated WRT (English dictionary coding) with
the PAQ6 data compressor. It compresses the Calgary corpus to 614,614
bytes in my tests. That is about 10K larger than PAQAR4, but 5 times
faster. I posted his code (open source and Windows .exe) on my web
page at
http://cs.fit.edu/~mmahoney/compression/#skibinski
I expect that WRT + PAQAR4 would set records on some benchmarks
(although dictionaries would not work for the Calgary challenge). My
tests (as 2 separate programs) produce 587,028 bytes. Unfortunately he
has not yet been able to integrate them. I think that PAQAR has some
interfacing between the model and archiver which make this more
difficult.
-- Matt Mahoney
|
|
