The documentation of the BernoulliMix program package has been written in Texinfo format, to make it possible to produce both on-line documentation and a printed manuals in dvi, ps and pdf formats easily. In the documentation, there is very little background theory on the models, rather references to the literature and available course books are given. It is stressed that the documentation is not designed to be any kind of replacement for a good text book on the topic. This underlines the role of the package as an auxiliary material to existing teaching material, not as a replacement to it.
In writing the documentation, we have concentrated on three aspects: the description of the usage of the programs, usage examples with functional descriptions, and the exercises. Each program is presented separately with the technical descriptions of command line options. Both long and short formats are available. The presentation of usage is followed by a handful of examples with the task described in the paragraph. This presentation ties together the actual execution of the programs and the machine learning modeling world taught in the classrooms. The examples are followed by exercises that form the core of the term project. The term project is to complete all the exercises. The easiest exercises follow closely the presented examples, but often the students are required to explain their findings more thoroughly. The most tedious one of the exercises is a model selection problem in modeling DNA copy number amplifications with a mixture model [Tikka et al., 2007,Myllykangas et al., 2008], where the student is asked to select an appropriate number of component distributions in a cross-validation setting by performing a 5-fold-cross-validation repeated 10 times for models ranging from 2 component distributions to 30.