Struct Random 

pub struct Random { /* private fields */ }

Random sparse wiring structure (unstructured sparsity).

Unlike NCP which has structured 4-layer connectivity, Random wiring simply removes a fraction of connections at random. This provides:

• Unstructured sparsity: No layer hierarchy
• Ablation baseline: Compare against structured sparsity
• Parameter reduction: Fewer synapses than fully-connected

When to Use

• Ablation studies: Test if NCP’s structure matters vs just sparsity
• Research: Compare structured vs unstructured sparse networks
• Simple sparse networks: When you don’t need biological structure

Example

use ncps::wirings::{Random, Wiring};

// Create random wiring with 50% sparsity
let mut wiring = Random::new(
    32,       // units
    Some(8),  // output_dim
    0.5,      // sparsity_level (50% connections removed)
    42,       // seed
);

wiring.build(16);

// Approximately half the possible synapses exist
let max_synapses = 32 * 32;  // 1024
let actual = wiring.synapse_count();
// actual ≈ 512 (with some randomness)

Sparsity Level

The sparsity_level parameter controls what fraction of connections to remove:

sparsity_level	Connections	Effect
0.0	100% (dense)	Same as FullyConnected
0.5	50%	Moderate sparsity
0.9	10%	Very sparse

Comparison with NCP

Aspect	Random	NCP
Structure	None	4-layer biological
Interpretability	Low	High
Information flow	Uncontrolled	Directed
Use case	Ablation	Production

Implementations

impl Random

pub fn new( units: usize, output_dim: Option<usize>, sparsity_level: f64, random_seed: u64, ) -> Self

pub fn from_config(config: WiringConfig) -> Self

Trait Implementations

impl Clone for Random

fn clone(&self) -> Random

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for Random

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

Formats the value using the given formatter.

impl Wiring for Random

fn units(&self) -> usize

Returns the total number of neurons (hidden units) in this wiring.

fn input_dim(&self) -> Option<usize>

Returns the input dimension (number of input features), or None if not yet built.

fn output_dim(&self) -> Option<usize>

Returns the output dimension (number of motor neurons).

fn build(&mut self, input_dim: usize)

Builds the wiring by setting the input dimension and creating sensory connections.

fn adjacency_matrix(&self) -> &Array2<i32>

Returns the internal adjacency matrix representing neuron-to-neuron synapses.

fn sensory_adjacency_matrix(&self) -> Option<&Array2<i32>>

Returns the sensory adjacency matrix (input-to-neuron connections).

fn add_synapse(&mut self, src: usize, dest: usize, polarity: i32)

Adds or modifies an internal synapse between two neurons.

fn add_sensory_synapse(&mut self, src: usize, dest: usize, polarity: i32)

Adds or modifies a sensory synapse from an input feature to a neuron.

fn get_config(&self) -> WiringConfig

Creates a serializable configuration for this wiring.

fn num_layers(&self) -> usize

Returns the number of logical layers in this wiring.

fn get_neurons_of_layer(&self, layer_id: usize) -> Vec<usize>

Returns the neuron IDs belonging to a specific layer.

fn is_built(&self) -> bool

Returns true if the wiring has been built (input dimension is set).

fn get_type_of_neuron(&self, neuron_id: usize) -> &'static str

Returns the type of a neuron by its ID.

fn erev_initializer(&self) -> Array2<i32>

Returns the reversal potential initializer (same as adjacency matrix).

fn sensory_erev_initializer(&self) -> Option<Array2<i32>>

Returns the sensory reversal potential initializer.

fn synapse_count(&self) -> usize

Returns the total number of internal synapses (non-zero entries in adjacency matrix).

fn sensory_synapse_count(&self) -> usize

Returns the total number of sensory synapses (input-to-neuron connections).

fn input_required(&self) -> bool

Returns true if this wiring requires external input (has sensory connections).