Hands-On Neuroevolution with Python

Genome fitness evaluation

Using the number of successful simulation steps returned by the run_cart_pole_simulation function described earlier, we are ready to implement the genome fitness evaluation function:

1. First, we run the cart-pole simulation loop, which returns the number of successful simulation steps:

    steps = run_cart_pole_simulation(net, max_bal_steps)

2. After that, we are ready to evaluate the fitness score of the particular genome, as described previously:

    log_steps = math.log(steps)    log_max_steps = math.log(max_bal_steps)    # The loss value is in range [0, 1]    error = (log_max_steps - log_steps) / log_max_steps    # The fitness value is a complement of the loss value    fitness = 1.0 - error

Refer to the eval_fitness(net, max_bal_steps=500000) ...