Trainers (Mountain Car, Tetris, and Helcopter)

Trainers are programs that put an agent into an environment and control the experiment.  Sample trainers are located in the /trainers directory.

There are two types of trainers: console trainers, and graphical trainers.

Console trainers are well suited to running a proper experiment, trying parameters, etc.  The graphical trainer gives a visual representation of the problem and can be very handy for debugging and visually evaluating your agent's performance.

We have provided several sample trainers, implemented in C/C++, Python, and Java.  They are all identical in function, and are located in:
/trainers/consoleTrainerJava
/trainers/consoleTrainerCPP
/trainers/consoleTrainerPython

These console trainers will run any of the Java domains, Mountain Car, Tetris, or Helicopter.  Check the source code in /trainers/<someTrainer>/src/ for an idea how to select which problem the trainer will select.  Special training programs are required for real time strategy and keep-away.

To run any trainer, go into its directory and type:
$>make
$>bash run.bash

Running a trainer is only half of what is necessary to run an experiment, you also need an agent!  See the previous section!
To rebuild a trainer, in that trainer's directory type:
$>make clean;make

The graphical trainer that will run for any of the Java domains is in:
/trainers/guiTrainerJava

To run this trainer, go into its directory and type:
$>bash run.bash

The source code for the GUI trainer is not provided.  This trainer is provided on an AS-IS basis.