def eval_genomes(genomes, config):
"""
The function to evaluate the fitness of each genome in
the genomes list.
The provided configuration is used to create feed-forward
neural network from each genome and after that created
the neural network evaluated in its ability to solve
XOR problem. As a result of this function execution, the
the fitness score of each genome updated to the newly
evaluated one.
Arguments:
genomes: The list of genomes from population in the
current generation
config: The configuration settings with algorithm
hyper-parameters
"""
for genome_id, genome in genomes:
genome.fitness = 0.0
net = neat.nn.FeedForwardNetwork.create(genome, config)
fitness = cart.eval_fitness(net)
if fitness >= config.fitness_threshold:
# do additional steps of evaluation with random initial states
# to make sure that we found stable control strategy rather than
# special case for particular initial state
success_runs = evaluate_best_net(net, config, additional_num_runs)
# adjust fitness
fitness = 1.0 - (additional_num_runs - success_runs) / \
additional_num_runs
genome.fitness = fitness
def evaluate_best_net(net, config, num_runs):
"""
The function to evaluate the ANN of the best genome in
specified number of sequetial runs. It is aimed to test it
against various random initial states that checks if it is
implementing stable control strategy or just a special case
for particular initial state.
Arguments:
net: The ANN to evaluate
config: The hyper-parameters configuration
num_runs: The number of sequential runs
Returns:
The number of succesful runs
"""
for run in range(num_runs):
fitness = cart.eval_fitness(net, max_bal_steps=additional_steps)
if fitness < config.fitness_threshold:
return run
return num_runs
def eval_genomes(genomes, config):
"""
The function to evaluate the fitness of each genome in
the genomes list.
Arguments:
genomes: The list of genomes from population in the
current generation
config: The configuration settings with algorithm
hyper-parameters
"""
for genome_id, genome in genomes:
genome.fitness = 0.0
net = neat.nn.FeedForwardNetwork.create(genome, config)
fitness = cart.eval_fitness(net)
if fitness >= config.fitness_threshold:
# do additional steps of evaluation with random initial states
# to make sure that we found stable control strategy rather than
# special case for particular initial state
success_runs = evaluate_best_net(net, config, additional_num_runs)
# adjust fitness
fitness = 1.0 - (additional_num_runs - success_runs) / \
additional_num_runs
genome.fitness = fitness
