What is this?
A simple game which looks a bit like a simplified version of Risk. It was implemented as the environment for building a player that learns through reinforcement learning. This project is an assignment for the Machine Learning course 2012.
ant compile in the root directory. This should build your classes in build/classes.
If things behave weird, or methods cannot be found, try recompiling by first running
ant clean and then again
Create a configuration file for each configuration you want to test. For example, if I want to test a different number of hidden units, create two files:
[network] hidden_units i 1
[network] hidden_units i 10
Now I can run these two experiments by calling:
java -cp build/classes TestGame ./default.conf ./nn-*.conf
Note that I first pass default.conf as this contains all the other configuration options. Only the parameters named in the individual experiment configurations will be altered for that particular experiment.
After all the experiments have finished, a summary will be printed. Also, for each run of each experiment a file named
config.conf-run-#.csv will be created, in which config.conf is replaced with the path of the configuration file for that experiment, and # is the number of the run.