Glossary

Algorithm

Algorithms listed in the software infopage are based on guesswork for compressors that are not open-source. Therefore the information provided should be considered with appropriate caution. Most of the time however the guesswork is likely to be correct.

Notable performance

The compressor has high rating performance in a test.

Disqualified

Compressors that are compatible with the elementary requirements (see submit compressor) will be tested in qualification rounds (see the qualification results at the results page). If the compressor performs below that of the current standard for qualifying, it will be disqualified and not tester any further, except for the Reference files test, where all compressors qualify.

If the compressor qualifies by performance, but fails verify checking (decompressed data does not match the original) it will be disqualified. If the reason for disqualifying is any other than performance, it is stated in the program page.

A special case for disqualification applies to BWT based compressors that freeze (practically for indefinite time) due to use of a simple string sorting.

Unrated

Unrated means that the compressor failed to qualify or that it failed error checking.

Qualification ratings

Ratings represent all the qualification rounds summed together. These are not comparable with other ratings. The purpose of these ratings is to be able to compare the relative performance of different software versions that did not qualify for the full tests. The decoder program size is excluded from these numbers to reduce random fluctuations that are caused by different versions having slightly different executable size.

GUI

Graphical User Interface

CLI

Command Line Interface

IPP

Integrated Performance Primitives. A library by Intel for multithreading and data processing.

SFX

Self-extracting archive. A feature in file archiver that produces an executable (like an installer program) that decompresses itself to produce the original files.

TAR

TAR is an archiver utility which stores files and metadata without compression into an archive file. Often used (as here) for single file compressors to produce a single output from multiple files.

BWT

Burrows Wheeler Transform. A method of transforming data before compression.

BWTS

Burrows Wheeler Transform variant that is inversible without prefix (32 bit space saving for 4 gigabyte file). Comparing BWT vs BWTS we note that BWTS does not increase compression ratio.

ST

Sort Transform, limited order Burrows Wheeler Transform.

LZ (or LZ77)

Lempel Ziv. A compression method that builds a dictionary and finds string matches from there.

LZP (or ROLZ)

A dictionary compression method which uses context to exclude some match offsets.

DICT

Dictionary build either dynamically by the compressor or statically linked into it. Usually limited to human language only. This is not related to the compression algorithms that use window called 'dictionary' to compress data.

X86

Relative to absolute address transformation for X86 executable files.

RLE

Run Length Encoding. Compresses repeated runs of symbols.

PPM

Prediction by Partial Match. A statistical compression method.

PPMII

PPM variant that inherits statistics for high orders from low orders.

HUF

Huffman entropy encoding, variable-length bit codes.

ARI

Arithmetic entropy coding. Encodes input into a single number.

IF (or SIF)

Inversion Frequencies. A method of encoding positions of each symbol in separate passes for each symbol (excluding known positions).

DC

Distance Coding. Like Inversion Frequencies, but all symbols encoded at once in a single pass.

MTF

Move To Front. A method of encoding symbol ranks (ranked by order of appearence) instead of the actual symbols.

WFC (or AWFC)

Weighted Frequency Count. A method of encoding symbol ranks (ranked by sliding window) instead of the actual symbols.

SSE (or APM)

Secondary Symbol Estimation (also known as Adaptive Probability Map) aims to improve a given probability (within a context) according to experience.

CM

Context Mixing. Mixing predictions of multiple models.