Hands-On Neuroevolution with Python

Genome encoding scheme implementation

The following source code implements ANN parameter estimations, as defined by the formula from the previous section (see the compute_weights_from_seeds function):

    idx = seeds[0]    theta = noise.get(idx, self.num_params).copy() * self.scale_by    for mutation in seeds[1:]:        idx, power = mutation        theta = self.compute_mutation(noise, theta, idx, power)    return theta

The compute_mutation function implements an estimation of the single step of the ANN parameter estimation, as follows:

    def compute_mutation(self, noise, parent_theta, idx, mutation_power):        return parent_theta + mutation_power * noise.get(idx,                                                         self.num_params)

The preceding code takes the vector of the parent's trainable parameters and adds to it ...