Enum set_genome::Mutations

pub enum Mutations {
    ChangeWeights {
        chance: f64,
        percent_perturbed: f64,
        standard_deviation: f64,
    },
    ChangeActivation {
        chance: f64,
        activation_pool: Vec<Activation>,
    },
    AddNode {
        chance: f64,
        activation_pool: Vec<Activation>,
    },
    AddConnection {
        chance: f64,
    },
    AddRecurrentConnection {
        chance: f64,
    },
    RemoveNode {
        chance: f64,
    },
    RemoveConnection {
        chance: f64,
    },
    RemoveRecurrentConnection {
        chance: f64,
    },
    DuplicateNode {
        chance: f64,
    },
}

Lists all possible mutations with their corresponding parameters.

Each mutation acts as a self-contained unit and has to be listed in the crate::Parameters::mutations field in order to take effect when calling [crate::Genome::mutate_with].

Variants

ChangeWeights

Fields

chance: f64

percent_perturbed: f64

standard_deviation: f64

See Mutations::change_weights.

ChangeActivation

Fields

chance: f64

activation_pool: Vec<Activation>

See Mutations::change_activation.

AddNode

Fields

chance: f64

activation_pool: Vec<Activation>

See Mutations::add_node.

AddConnection

Fields

chance: f64

See Mutations::add_connection.

AddRecurrentConnection

Fields

chance: f64

See Mutations::add_recurrent_connection.

RemoveNode

Fields

chance: f64

See Mutations::remove_node.

RemoveConnection

Fields

chance: f64

See Mutations::remove_connection.

RemoveRecurrentConnection

Fields

chance: f64

See Mutations::remove_recurrent_connection.

DuplicateNode

Fields

chance: f64

See Mutations::duplicate_node.

Implementations

impl Mutations

pub fn add_connection(genome: &mut Genome, rng: &mut impl Rng) -> MutationResult

This mutation adds a new feed-forward connection to the genome, should it be possible. It is possible when any two nodes¹ are not yet connected with a feed-forward connection.

---

1. “any two nodes” is technically not correct as the start node for the connection has to come from the intersection of input and hidden nodes and the end node has to come from the intersection of the hidden and output nodes. ↩

impl Mutations

pub fn add_node( activation_pool: &[Activation], genome: &mut Genome, rng: &mut impl Rng )

This mutation adds a new node to the genome by “splitting” an existing connection, i.e. the existing connection gets “re-routed” via the new node and the weight of the split connection is set to zero. The connection leading into the new node is of weight 1.0 and the connection originating from the new node has the same weight as the split connection (before it is zeroed).

impl Mutations

pub fn add_recurrent_connection( genome: &mut Genome, rng: &mut impl Rng ) -> MutationResult

This mutation adds a recurrent connection to the genome when possible. It is possible when any two nodes ¹ are not yet connected with a recurrent connection.

---

1. “any two nodes” is technically not correct as the end node has to come from the intersection of the hidden and output nodes. ↩

impl Mutations

pub fn change_activation( activation_pool: &[Activation], genome: &mut Genome, rng: &mut impl Rng )

This mutation changes the activation function of one random hidden node to any other choosen from activation_pool. If the pool is empty (the current activation function is excluded) nothing is changed.

impl Mutations

pub fn change_weights( percent_perturbed: f64, standard_deviation: f64, genome: &mut Genome, rng: &mut impl Rng )

This mutation alters percent_perturbed connection weights sampled from a gaussian distribution with given standard_deviation.

impl Mutations

pub fn duplicate_node( genome: &mut Genome, rng: &mut impl Rng ) -> Result<(), MutationError>

This mutation adds a new node to the genome by “duplicating” an existing node, i.e. all incoming and outgoing connections to that node are duplicated as well.

The weight of all outgoing connections is half the initial weight so the mutation starts out functionally equivalent to the genome without the mutation. By duplicating a node and its connections small local symetry can develop. It draws inspiration from the concept of gene duplication and cell division.

impl Mutations

pub fn remove_connection( genome: &mut Genome, rng: &mut impl Rng ) -> MutationResult

Removes a connection, should this be possible without introducing dangling structure. Dangling means the in- or out-degree of any hidden node is zero, i.e. it neither can receive nor propagate a signal. If it is not possible, no connection will be removed.

impl Mutations

pub fn remove_node(genome: &mut Genome, rng: &mut impl Rng) -> MutationResult

Removes a node and all incoming and outgoing connections, should this be possible without introducing dangling structure. Dangling means the in- or out-degree of any hidden node is zero, i.e. it neither can receive nor propagate a signal. If it is not possible, no node will be removed.

impl Mutations

pub fn remove_recurrent_connection( genome: &mut Genome, rng: &mut impl Rng ) -> MutationResult

Removes a recurrent connection if at least one is present in the genome. Does nothing when no recurrent connections exist.

impl Mutations

pub fn mutate(&self, genome: &mut Genome, rng: &mut impl Rng) -> MutationResult

Mutate a Genome but respects the associate chance field of the Mutations enum variants. The user needs to supply some RNG manually when using this method directly. Use crate::Genome::mutate as the default API.

Trait Implementations

impl Clone for Mutations

fn clone(&self) -> Mutations

Returns a copy of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for Mutations

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for Mutations

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for Mutations

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>where __S: Serializer,

Serialize this value into the given Serde serializer.