[mlpack] Neural Evolution Algorithms - Week 10

[bang liu]

Dear all,

Last week, our NEAT algorithm has passed the Cart Pole balancing problem.
Besides, we implemented the Mountain Car testing problem and it also passed
(only need 1 generation to evolve for 500 genome population, and need
around 10 generations for 10 genome population size).
Marcus helped to implement the test case of Super Mario game.

Besides, I am implementing the double pole balancing problem. Currently I
have been clear about the formulations of double pole task, and have
implemented the key part in double pole balancing: using Runge-Kutta 4th
order methods to update 6 states. I am trying to organizing the remaining
part (evaluate the fitness of a genome) to make it as clear as possible, as
current codes I referred to seems not quite neat.

The materials I reffered to for double pole balancing are:
https://github.com/CodeReclaimers/neat-python/tree/mstechly-master/examples/pole_balancing/double_pole
Evolving neural network controllers for unstable systems by A. P. Wieland
et. al (this paper contains the formulations of double pole balancing)
https://en.wikipedia.org/wiki/Runge–Kutta_methods

After NEAT pass the double pole and Mario game, we will clean the code to
merge CNE and NEAT to mlpack, and start implementing HyperNEAT.

Best,
Bang
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cc.gatech.edu/pipermail/mlpack/attachments/20160726/44eb3532/attachment-0002.html>