Crate ncps_rust 

NCPS - Neural Circuit Policies (Rust)

Port of the Python NCPS library to Rust using the Burn framework.

Features

• LTC: Liquid Time-Constant RNN cells with ODE-based dynamics
• CfC: Closed-form Continuous-time cells (3 modes: default, pure, no_gate)
• NCP: Neural Circuit Policy wiring (biologically-inspired sparse connectivity)
• Sparsity Masks: Wiring adjacency properly enforced in forward passes
• Input/Output Mapping: Affine, Linear, or pass-through modes
• Mixed Memory: LSTM augmentation for long-term dependencies

Quick Start

use ncps::prelude::*;

// Create a wiring configuration
let mut wiring = AutoNCP::new(32, 8, 0.5, 22222);
wiring.build(16); // 16 input features

assert_eq!(wiring.units(), 32);
assert_eq!(wiring.output_dim(), Some(8));

Cell-level Usage

For direct cell access (single timestep processing):

use ncps::cells::{LTCCell, MappingMode};
use ncps::wirings::FullyConnected;

let wiring = FullyConnected::new(32, Some(8), 1234, true);
let cell = LTCCell::<Backend>::new(&wiring, Some(16), &device)
    .with_input_mapping(MappingMode::Affine, &device);

Modules

activation

Custom activation functions for NCPS

cells

prelude

rnn

RNN Layer Modules

wirings