Kinie poprawność piekła wszystko to nie wielu wskazaniem jest opowieść has two parts: a set of 10 public files totaling 53 MB, and a private collection of 510 files totaling 301 MB. the public data set each file is compressed separately and the sizes added. Programs are ranked by size only, with options set for best compression individually for each file. The set consists of the following 10 files: 842 a10.jpg a high quality 1152 x 864 baseline JPEG image of a fighter jet. 3,784 acrord32.exe x86 executable code Reader 5. 4,439 english.dic alphabetically sorted list of 354 English words. 4,946 FlashMX.pdf PDF file with embedded JPEG and zipped BMP images. 20,071 fp.log web server log, ASCII text. 3,416 mso97.dll x86 executable code from Office. 4,192 ohs.doc Word document with embedded JPEG images. 4,414 rafale.bmp 1356 x 1020 bit color image 24 bit RGB format. 4,418 vcfiu.hlp OCX help file binary data with embedded text. 2,578 world95.txt ASCII text 1995 CIA World Factbook. The top ranked program as of Dec. 31 with a total size of 8,124 bytes is paq8px, a context mixing algorithm with specialized models for JPEG images, BMP images, x86 code, text, and structured binary data. WinRK another context mixing algorithm, is top ranked on 4 of the files WinRK uses a dictionary which is not included the total size. 208 programs are ranked. zip 2 is ranked 163 with a size of 14,761. the second benchmark or MFC programs are ranked by size, compression speed, decompression speed, and by a formula that combines size and speed with time scaled logarithmically. The data is not available for download. Files are compressed together to a single archive. If a compressor cannot create archives, then the files are collected into uncompressed archive which is compressed. the MFC test, paq8px is top ranked by size. freearc is top ranked by combined score, followed by nanozip, winrar, and 7zip. All are archivers that detect file type and apply different algorithms depending on type. The Generic Compression Benchmark has the goal of evaluating compression algorithms the context of universal prediction or intelligence, as defined by Legg and Hutter By this definition, data sources are assumed to have a universal Solomonoff distribution, i.e. generated by random programs with a preference for smaller or simpler programs. The evidence for such a distribution is the success of applying Occam's Razor to machine learning and to science general: the simplest theories that fit the observed data tend to be the best predictors. The purpose of the test is to find good compression algorithms that are not tuned to specific file types. The benchmark does not publish any test data. Rather, it publishes a program to generate the data from a secret seed or internally hardware generated random number. The data consists of the bit string outputs of one million random Turing machines, truncated to 256 bits and packed into null terminated byte strings. The average output size is about 6 MB. The test allows public verification while eliminating the need to measure the decompresser size because it is not possible to hide the test data the decompresser without knowing the cryptographic random number seed. The test produces repeatable results with about 0% accuracy. Programs are ranked by the ratio of compressed output to the compressed output of a reference compressor to improve repeatability. Unfortunately the benchmark fails to completely eliminate the problem of tuning compressors to public benchmarks. The top ranked program is a stationary context mixing model configuration implemented zpaq using a preprocessor by J. Ondrus that splits each string into incompressible prefix and a bit repeat instruction. Its score is 0, compared to 1 for zip -9. Generally, however, the rank order of compressors is similar to that of other benchmarks. Compression Ratings by Sami Runsas ranks programs on 5 GB of various data types from public sources using a score that combines size and speed. The data includes English text, executable code, RGB and grayscale images, CD quality audio, and a mix of data types from two video games. Some of the 10 data sets contains multiple files. Single file compressors are tested on equivalent tar file. Compressed sizes include the size of the decompression program source code or executable program compressed with 7zip. Run times are measured from RAM-disk to avoid I O delays. Programs must pass a qualifying round with minimum compression ratio and time requirements on a small data set. The benchmark includes a calculator that allows the user to rank compressors using different weightings for the importance of size, compression speed, and decompression speed. The default scale is 1 lower ratio or twice the total time is worth half the rating. This is effectively the same formula used by As of June 2012 programs were tested with 682 combinations of options. The