Author: Ratko V Tomic
Date: 19:46:00 01/03/06
Go up one level in this thread
On January 03, 2006 at 13:53:24, Dann Corbit wrote: >This is something really incredible! > >Have you made a post to news:comp.compression to tell them about it? > There are couple posts (in comp.compression & sci.math), one before one after the release of the source code. The last one is at: "Quantized Indexing Source Code Available" http://groups.google.com/group/comp.compression/browse_thread/thread/7053c23c0d01c81c and that links back to the earlier discussion. The unmoderated usenet groups are unfortunately a bit of a jungle, with all kinds of nonsense mixed up with few good comments. One good side-effect was getting in touch with few serious guys and continuing discussion via email. >If I interpret the paper correctly, your algorithm could also be used to >generate a blazing-fast perfect hash function. That is something that occured to me as well, but I never got to try it in that role, as a perfect hash generator. There are just too many possibilities that open up with a highly practical, yet "optimal" (the tightest code at any given arithmetic precision) enumeration algorithm, so I put out the source code for others to explore. The perfect hash is most similar to the field of constrained coding, for which QI has a nice formalism and recurrences with factored out constraints. On page 2 in arXiv preprint [T3] I mentioned Immink's work, that's the place to check first on highly constrained coding: Immink: http://www.exp-math.uni-essen.de/~immink/refs.html Janssen: http://www.mat.univie.ac.at/~nuhag/Janssen/ With perfect hash, the constraint is the presence of entry in the dictionary. Even the conventional EC, which is O(n) times slower and needs O(n) larger coding tables than QI, was quite useful in constrained coding where it goes back for decades (it is used for low level media recording codes). The new algorithm will give them huge performance boost. It's an interesting area to explore. >I am simply incredulous! Many are on just hearing the performance claims. Compression field has more than its share of frauds and scam artists with grand claims about secret "algorithms" (usually a well hidden communication link to another computer holding the uncompressed data). That's why I diverted myself from work and from more enjoyable explorations of the algorithm for several weeks, to write the preprint & to clean up couple tech reports, and hurried up the source code for relase (which still needs more cleanup & more functions ported into from the large and messy research prototype program that I use, which goes back to the that 2AM 32-bit input version).
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.