Class Connection<T>

Namespace

AiDotNet.NeuralNetworks

Assembly

AiDotNet.dll

Represents a connection between two nodes in a neural network, particularly used in evolving neural networks.

public class Connection<T>

Type Parameters

T

The numeric type used for the weight, typically float or double.

Inheritance

object

Connection<T>

Inherited Members

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

object.ToString()

Remarks

A Connection represents a weighted link between two nodes in a neural network. Connections are fundamental elements in neural networks, allowing signals to flow from one node to another with a specific weight. This class is particularly designed for use in evolutionary algorithms like NEAT (NeuroEvolution of Augmenting Topologies), which evolve both the weights and structure of neural networks. The Innovation number serves as a historical marker to track the evolutionary lineage of connections.

For Beginners: A Connection is like a wire connecting two parts of a neural network.

Think of a neural network as a system of connected nodes (like neurons in a brain):

• Each Connection is like a wire that passes signals from one node to another
• The Weight determines how strong the signal is (like a volume knob)
• IsEnabled acts like an on/off switch for the connection
• The Innovation number is like a birth certificate that shows when this connection first appeared

For example, if node 3 connects to node 5 with a weight of 0.7, signals from node 3 will reach node 5, but their strength will be multiplied by 0.7 (either amplified or reduced depending on the original value).

Constructors

Connection(int, int, T, bool, int)

Initializes a new instance of the Connection<T> class with the specified parameters.

public Connection(int fromNode, int toNode, T weight, bool isEnabled, int innovation)

Parameters

fromNode int

The identifier of the source node.

toNode int

The identifier of the target node.

weight T

The weight of the connection.

isEnabled bool

A value indicating whether the connection is enabled.

innovation int

The innovation number of the connection.

Remarks

This constructor creates a new Connection with the specified parameters. It sets up all properties of the connection, including which nodes it connects, its weight, whether it's enabled, and its innovation number for evolutionary tracking.

For Beginners: This is the "recipe" for creating a new connection between two nodes.

When creating a new connection, you need to specify:

• fromNode: The ID of the starting node
• toNode: The ID of the ending node
• weight: How strong the connection is
• isEnabled: Whether the connection is turned on
• innovation: The unique ID number for this connection type

For example, new Connection(3, 5, 0.7, true, 42) creates a connection:

• From node 3 to node 5
• With a weight of 0.7
• That is currently enabled (active)
• With innovation number 42 (tracking its evolutionary origin)

Properties

FromNode

Gets the identifier of the node from which the connection originates.

public int FromNode { get; }

Property Value

int

The identifier of the source node.

Remarks

The FromNode property identifies the node where the connection begins. In the context of signal flow, this is the source node that sends a signal through the connection to the target node.

For Beginners: This is the ID number of the starting point of the connection.

Think of it like the address of the house where a mail delivery begins:

• Each node in the network has a unique ID number (like a house address)
• The FromNode tells you where the signal starts its journey
• In a diagram, this would be the node on the left side of an arrow

For example, if FromNode is 3, the connection starts at node #3 in the network.

Innovation

Gets the innovation number, a historical marker that uniquely identifies the connection in the context of evolution.

public int Innovation { get; }

Property Value

int

The innovation number of the connection.

Remarks

The Innovation property is a unique identifier assigned to each new connection during the evolution process. It serves as a historical marker that helps track the origin of connections across different generations and individuals. This is crucial in evolutionary algorithms like NEAT, where matching genes (connections) across individuals is necessary for effective crossover operations.

For Beginners: The Innovation number is like a birth certificate or ID for the connection.

Think of it as a timestamp or serial number:

• Each time a new type of connection appears during evolution, it gets a unique Innovation number
• Two connections with the same Innovation number in different networks represent the same "genetic innovation"
• This helps when combining two networks (like parents having a child):
  • If both parent networks have a connection with Innovation #42, the child will inherit one of them
  • If only one parent has a connection with Innovation #86, the child might inherit that unique connection

This tracking system helps maintain diversity while allowing networks to share successful innovations.

IsEnabled

Gets or sets a value indicating whether the connection is enabled and actively transmitting signals.

public bool IsEnabled { get; set; }

Property Value

bool

True if the connection is enabled; otherwise, false.

Remarks

The IsEnabled property determines whether the connection actively participates in signal transmission. When disabled, the connection is still part of the network structure but does not pass any signal. This is particularly useful in evolutionary algorithms, where disabling connections (rather than removing them) allows for potential reactivation in future generations.

For Beginners: IsEnabled works like an on/off switch for the connection.

Think of it like a light switch:

• When IsEnabled is true, the connection is active and passes signals
• When IsEnabled is false, the connection is inactive and blocks signals

This feature is especially useful in evolving neural networks because:

• It allows the network to "turn off" connections that aren't helpful
• The connection can be turned back on later if needed
• This is often better than completely removing the connection

For example, during evolution, a connection might be disabled if it leads to poor performance, but could be re-enabled in a later generation if the network's structure changes.

ToNode

Gets the identifier of the node to which the connection leads.

public int ToNode { get; }

Property Value

int

The identifier of the target node.

Remarks

The ToNode property identifies the node where the connection ends. In the context of signal flow, this is the target node that receives a signal from the source node through the connection.

For Beginners: This is the ID number of the end point of the connection.

Think of it like the destination address for a mail delivery:

• The ToNode tells you where the signal ends its journey
• In a diagram, this would be the node on the right side of an arrow

For example, if ToNode is 5, the connection ends at node #5 in the network.

Weight

Gets or sets the weight of the connection, which determines the strength of the signal transmission.

public T Weight { get; set; }

Property Value

T

The weight of the connection.

Remarks

The Weight property determines how strongly the signal from the source node influences the target node. Positive weights enhance the signal, negative weights inhibit it, and the magnitude indicates the strength of the effect. During learning or evolution, weights are adjusted to improve the network's performance.

For Beginners: The Weight is like a volume knob for the signal passing through the connection.

Think of the weight as a multiplier:

• A positive weight (like 0.7) passes the signal through, but might reduce its strength
• A negative weight (like -0.3) reverses the signal (if it was "turn right", it becomes "turn left")
• A weight with larger magnitude (like 2.5) amplifies the signal, making it more influential
• A weight near zero (like 0.01) makes the signal very weak, almost ignoring it

When a neural network learns, it's mostly adjusting these weight values to get better results.