Students are expected to work in groups of two or three throughout the project. To pass the course, the group must

1. Turn in solutions to four programming tasks, due on weeks 2, 3, 4 and 6 of the course (see below for details). Each solution is a tar file which when unpacked includes a 'run.sh' script that produces all data that have been assigned so far. These packages have to be delivered by tuesday, 9 o'clock by mail (wettig@cs.helsinki.fi) such that they can be checked in time before class.
2. Give a short presentation about their solutions to each programming task to other students
3. Turn in a written final report covering all four problems.

All students in the same group receive the same grade for the course. The grade of a group is based on

• Written end report (65% of grade)
• Compression performance calculated from the sizes of the group's solutions to programming tasks, 1–4 (35% of grade).

The written report should show that the students are able to choose appropriate tools for compressing given data and can justify their choices. Where you choose to use readily available software, you must explain what it does and why this is appropriate for the task at hand.

For the first programming task, the data will be simple and students will be given the model from which is has been generated. The purpose of this task is to make sure that students get a good start and build up some tools they can use on later tasks where finding appropriate models is an essential part of the task. Therefore, for the first tasks students are required to use arithmetic coding. It is allowed (and even recommended) that you use an existing implementation of arithmetic coding, but you must make sure you understand how it works.

The second data set will be a simple time series. The third data set will be from a classification problem. The fourth data set will consist of image data.

You are allowed to use existing software packages, but in your written report you are required to explain the compression algorithms they are based on. You are also required to explain why you think a particular package is a better choice than using some of the methods you learned about in the course. A good explanation here would be based on actually trying something to find out how it works. We try to choose the data sets in such a way that it is possible to beat the performance of off-the-shelf packages by using coding algorithms we know from the course together with some relatively simple models of the data.

Results

There have been three groups. The package sizes in bytes (in case of problem 5 the error penalty) are listed below. The final score is a weighted sum of these numbers, namely 0.1*x₁ + 0.2*x₂ + 0.3*x₃ + 0.4*x_4/5 , where x_i is the package size of the package for problem i. The final reports can also be found below.