Automatic parallelized hyperparameter search with NEAT Reinforcement Learning using OpenAI Gym and Hyperopt

Testing a hyperparameter search with parallelized implementation of the NEAT algorithm (NeuroEvolution of Augmenting Topologies) on some OpenAI gym environments

The example is one of the simplest possible implementation using a modified version of the neat-python library and the Humanoid-v3 environment.

If you prefer to use the original version of neat-python you have to delete the following line out of the config file (which will repopulate a species with the best genome after total extinction)

reset_with_best       = 1

Then the example can simply be run by typing (it will start an automated search)

python run_this.py

If you want to just run the config file (and no search) you need to comment out everything in the "__main__" part except of the last line.

After 570 (just using the config file / w.o. hyperparameter search)