2nd Semester
The second semester still was not much about creating an actual product. Instead we had a lot of lessons about how technology affects peoples life, and we had to considder whether our project would actually do anyone anything good and more of such things.
Courses and grades
| Course Title | Exam form | Grade |
| Mathematics | oral | passed |
| C Programming | oral | passed |
| Computer supported calculations | oral | passed |
Project grade: 9
Project: Modelling of probability in tænkeboks using bayesian networks
We decided to do a project on how to calculate the likelihood of dice rolls in the drinking game tænkeboks, using bayesian networks.
In tænkeboks everyone gets 4 dice, which he rolls with a dice cup. Each playyer may now peek at his dice and see what he got. The goal is to predict how many dices of a kind there are all in all. The first player start with making a guess on how many dice of a kind there are. The next player can chose not to belief his guess, or he can go farther by either increasing the number of dices in the guess or the kind of dices. If someone does not belief the guess, he may ask to see all dices. If the guess was correct (and there were at least that many of that kind) he loses one dice, otherwise the preceding player loses one dice. THis goes on until there is only one player left.
Bayesian networks are used to model unmcertanity. A bayesian network is a graph in which the nodes represent the events that should be modelled. Edges connecting the nodes represent dependencies. That means that if two nodes are connected by an edge, the probability for the outcome in one of the nodes depends on the outcome of the event in the other node. These dependencies can (theoretically) span multiple levels, this way complex situations can be modelled. Using the Bayes formula it is possible to calculate the probability for all events in the graph, and the numbers get close to reality, the more input the bayesian network gets (i.e. how many known event outcomes are specified).
For example, given a number of symptoms the right bayesian network would be able to calculate the probability for a specific illness a patient might have (provided with the probability for certain symptons und certain conditions for certain illnesses, and of course the symptons the patient shows).
During the project we were able to create a bayesian network that was able to calculate the probability for various dice rolls for up to 2 dice for two players. We were using HUGIN to create the network, though in the end we had to do a lot of manual work. Adding more dices resultet in the exponential increase of memory usage (more than 1 GB), which is why we were only able to model 4 dices all in all.
We learned a lot about bayesian networks this semester, which is cool. We even got to hear that the technical level of the report was on par with that of a last semester project, the rest of the report was not so good though. All in all this was an ok project, but none too interesting really.
Last updated: 2005-11-25
