« Hufint » : différence entre les versions
Aller à la navigation
Aller à la recherche
Contenu ajouté Contenu supprimé
Aucun résumé des modifications |
Aucun résumé des modifications |
||
Ligne 35 : | Ligne 35 : | ||
The algorithm is rather starightforward but not documented here. |
The algorithm is rather starightforward but not documented here. |
||
At least two public functions (in a java lib) are defined: |
|||
<pre>int f(BitSet) /* conversion of bitSet into a positive integer - decoding */ |
<pre>int f(BitSet) /* conversion of bitSet into a positive integer - decoding */ |
||
BitSet f(int) /* conversion of a positive integer into bitSet - encoding */ </pre> |
BitSet f(int) /* conversion of a positive integer into bitSet - encoding */ </pre> |
||
Similar functions are available for arrays of integers.<br>Various other convenience functions aredefined for reporting and checking.<br> |
Similar functions are available for arrays of integers.<br>Various other convenience functions aredefined for reporting and checking.<br> |
||
<br> |
<br> |
Version du 22 janvier 2014 à 09:17
Purpose
A Hufint is a variable-length-coded integer, with an optimal compression perspective.
Hufint stands for Huffman integer, or Huffman coded integer.
The contract of this coding tool is:
- allow to encode any positive integer (including 0) a a set of bits
- allow a non-ambiguous decoding algorithm
- use less bits for smaller integer
- allow non-ambiguous encoding/decoding for arrays of integers.
Hufint are used in the Matscape projects, wherever
- integer values are likely to be small (example: number of child)
- integer values are allowed to be large (but rather seldom)
- integer statistical distribution is roughly an exponential decrease
- encoding size is critical
- speed for math operations on the values is not critical
Hufint base
The Hufint base is the slice size for the Hufint bitset representation.
Matscape uses Hufint-6 representations (Hufint base = 6), where
- numbers from 0 to 2^5-1 (31) are represented on 6 bits
- numbers from 2^5 (32) to 2^10-1 (1023) are represented on 12 bits
- numbers from 2^10 (1024) to 2^15-1 (1023) are represented on 18 bits
- etc...
Algorithm
The algorithm is rather starightforward but not documented here.
At least two public functions (in a java lib) are defined:
int f(BitSet) /* conversion of bitSet into a positive integer - decoding */ BitSet f(int) /* conversion of a positive integer into bitSet - encoding */
Similar functions are available for arrays of integers.
Various other convenience functions aredefined for reporting and checking.