December 2019
Intermediate to advanced
368 pages
11h 10m
English
The source code implementing the evaluation of the genome fitness is similar to the one used for a single-pole balancing experiment. The main difference is that it will refer to another simulation environment to get the number of balanced steps. Thus, you can refer to the source code of the eval_fitness(net, max_bal_steps=100000) and eval_genomes(genomes, config) functions in the two_pole_markov_experiment.py file for implementation details.
In this experiment, we introduced adaptive learning, which will try to find the correct short pole length during the evolution process. The length of the short pole changes the motion dynamics of the system. Not all combinations of hyperparameters combined with a particular ...
Read now
Unlock full access