Semesterwork (6th) in simulation technics was to implement an agent for playing the GO game using a neuronal net. We had a teamsize of 3 students.
package smartness.*

My part was to implement an object-oriented neuronal-net.
It was about a feedforward net which is activated by the logistic function and trained by the backpropagation algorithm with the batch-method. The topologie, the teaching patterns as well as the weights between the neurons are persistent in propertyfiles.

A basic experiment was learning the numbers from 0-9, a second one was learning the alphabet. It was not so easy to find a net-topology to learn with.

Our player worked well because we didn´t teach him just pattern of GO, remember even if u reduce the size of the playfield to 9×9 it is still 3 to the power of 81 posibilities, which is much more than u can teach in a lifetime. The player is rating the opponent´s move. The agent reduces the problem by tracking just a small space on the field and rotating this space. Our agent won against all other agents in a contest :)