A Robocode Robot with neural network function approaximation.
The robot can choose to either do wave surfing, move forward/backward or fire based on the current states.
The neural network is now built on Encog for better learning performance.

Training against most bots can be done in 1000 rounds.
When evaluating the performance after training, better set exploration off in the source code (epsilon = 0.0;) for optimal performance.

Against TrackFire: ~99% win rate.
Against SpinBot: ~100% win rate.
Against Tracker: ~99% win rate.
Against Fire: ~100% win rate.
...
...

1. Add Encog (encog.org) to dependency.
2. Compile project.
3. Run project with configuration "-Dsun.io.useCanonCaches=false -Ddebug=false -DNOSECURITY=true".
4. In "Preferences/Development Options", add "$PROJECTFOLDER/out/production/Robocode_MLProject".
5. In "Battle/New Battle" select "bots.BasicWaveSurferBot" and any opponent to start training.