What steps will reproduce the problem?
1. Start up the Nero mod and change the stick together and/or approach enemy
slider to a nonzero value.
2. Spawn a turret and some agents.
3. Read the output and see that the champion agent has a fitness of 0
line 305 of NeroEnvironment.py
What is the expected output? What do you see instead?
The expected output is for the champion agent to have a nonzero fitness and for
some learning to take place. Instead, the champion has a zero fitness and no
discernible learning takes place.
What version of the product are you using? On what operating system?
1312 on a machine in the CS labs.
Please provide any additional information below.
We found the source of the problem in NeroEnvironment.py, line 305. It
mistakenly uses
R[constants.FITNESS_STAND_GROUND] = -d*d
instead of
R[constants.FITNESS_STICK_TOGETHER] = -d*d
To assign the reward for sticking together. Fixing this fixes the bug.
Original issue reported on code.google.com by McHob...@gmail.com on 3 Nov 2011 at 8:40
Original issue reported on code.google.com by
McHob...@gmail.com
on 3 Nov 2011 at 8:40