m1

The M1 project was intended to improve context mixing and to make it more practical, while transfering common sense procedures from engineering. In the field of compression parameter optimization and model fitting to training data seems to be completely new. The redundancy contained implicitly in compression software via ad-hoc parametrization is often ignored. M1 has proven that this redundancy can be significant. Another issue addressed by M1 is, that context mixing can be implemented more efficient than paq, i.e. a significant speedup can be obtained compared to an *equivalent* paq model layout (e.g. lpaq modification).

All M1 builds require parameter files to run. These files store the configuration of the statistical model, encoded as bit vectors in a human readable form. Precreated profiles reside within the profile/ directory.

Optimizing the parameters of the statistical model can be done via invoking the optimizer build,...

The M1x2 0.6 program was submitted in March 23, 2010 with instructions to use the enwik7.txt text profile: "text is a fairly good general purpose optimization target". The program version is presented as x2-0.6 to associate it as a part of the "M1 project".

(Have info that should be added here? E-mail.)