Hawkynt

PAQ (Context Mixing)

Reduced lpaq-style context-mixing primitive: six hashed bit models over byte orders 0, 1, 2, 3, 4 and 6, blended by a single logistic-domain mixer trained by online gradient descent, refined by one adaptive probability map (SSE) keyed on the previous byte, and entropy-coded with a 30-bit binary arithmetic coder. Byte-for-byte identical to CompressionWorkbench’s BB_ContextMixing reference block. This is deliberately NOT the full PAQ ensemble - the real PAQ8/cmix model sets add word, sparse, indirect, record and media-specific models behind a much larger mixing network, none of which is implemented here.

Properties

Property	Value
Category	Compression Algorithms
Sub-category	Context Mixing
Security status	🎓 Educational Only
Complexity	Advanced
Inventor	Matt Mahoney (context mixing / PAQ family)
Year	2002
Origin	🇺🇸 United States
Source	algorithms/compression/paq.js

Security

Status: 🎓 Educational Only

No vulnerabilities are recorded for this implementation.

Documentation

• PAQ - Wikipedia
• Data Compression Explained (context mixing, SSE)
• PAQ Data Compression Programs

References

• Context Mixing - Wikipedia
• Hutter Prize Competition
• Adaptive Weighing of Context Models (Mahoney, 2005)

Test vectors

5 vectors ship with this algorithm and run in the test suite. Byte values are hexadecimal.

Vector 1 — Empty input - header only

Field	Value
input	(empty)
expected	00000000

Vector 2 — Single byte 0x41

Field	Value
input	41
expected	010000004180

Vector 3 — Repeated English pangram (4x)

Field	Value
input	74686520717569636b2062726f776e20 666f78206a756d7073206f7665722074 6865206c617a7920646f672e20746865 20717569636b2062726f776e20666f78 206a756d7073206f7665722074686520 6c617a7920646f672e20746865207175 69636b2062726f776e20666f78206a75 6d7073206f76657220746865206c617a 7920646f672e2074686520717569636b 2062726f776e20666f78206a756d7073 206f76657220746865206c617a792064 6f672e20
expected	b4000000746fba2ff6bc8eb814ab0b9e 20c6ccb67bbd62993cf549b066f53fca 1066c11768c6804aeec60b4a3c62

Vector 4 — 256 identical bytes

Field	Value
input	61616161616161616161616161616161 61616161616161616161616161616161 61616161616161616161616161616161 61616161616161616161616161616161 61616161616161616161616161616161 61616161616161616161616161616161 61616161616161616161616161616161 61616161616161616161616161616161 61616161616161616161616161616161 61616161616161616161616161616161 61616161616161616161616161616161 61616161616161616161616161616161 61616161616161616161616161616161 61616161616161616161616161616161 61616161616161616161616161616161 61616161616161616161616161616161
expected	0001000061a9bdf4

Vector 5 — All 256 byte values in order

Field	Value
input	000102030405060708090a0b0c0d0e0f 101112131415161718191a1b1c1d1e1f 202122232425262728292a2b2c2d2e2f 303132333435363738393a3b3c3d3e3f 404142434445464748494a4b4c4d4e4f 505152535455565758595a5b5c5d5e5f 606162636465666768696a6b6c6d6e6f 707172737475767778797a7b7c7d7e7f 808182838485868788898a8b8c8d8e8f 909192939495969798999a9b9c9d9e9f a0a1a2a3a4a5a6a7a8a9aaabacadaeaf b0b1b2b3b4b5b6b7b8b9babbbcbdbebf c0c1c2c3c4c5c6c7c8c9cacbcccdcecf d0d1d2d3d4d5d6d7d8d9dadbdcdddedf e0e1e2e3e4e5e6e7e8e9eaebecedeeef f0f1f2f3f4f5f6f7f8f9fafbfcfdfeff
expected	000100000023548f83aafcd049612c23 52f8e7db2066137c77b4546afe35df5f 629eb0a8b5ab0a6c7f8e874c25e1ffec cbcd7859900afdd89194ab15ee00192d cee052025bbc2a78214b6749aa37ba25 6066e0170d57fb42043e1637a519b8de f70a0268833b6b0662f34d81e2822b1b d4da4e171da20ae235dfe18d7c833f33 a54022db5be2dc44ed5e1f9564a10527 dd6aed4ee432d5351a558e47b92f9452 ba1fd41943348b4d58e538923f8329bd d55ae0735d0f823c967fa26abec72ad0 842d21ab95090142e9a5105d4b4394ea 425f56d5993bafc236d69426f52da539 5c9925f9541f03711b604299cb70792d 6a3bd08c

---

← All algorithms

This site is open source. Improve this page.