walnutjs-2.1

Jaap Murre and Jurriaan Schreuder, University of Amsterdam, 2019-2024

A framework for neurocognitive neural network modelling in JavaScript with a 37-year history.

Approach

Contrary to most other neurosimulators, including the predecessors to this version, Walnut 2.1 aims to have a stricter division of labor between:

• neural network architecture
• neural algorithms (colled in so called neural paradigms)
• display for visualization and interaction

History: Walnut neurosimulator, Walnut Nutshell, Walnut.py, and WalnutJS 1.0

Predecessors to Walnut were developed in C by Jaap Murre at Leiden University from 1988 to 1992, including the Metanet neurosimulator with Steven Kleynenberg.

The Walnut neurosimulator with the Walnut Nutshell interactive graphical user-interface was developed in C++ in the late 90s by Jaap Murre at the NeuroMod group (still downloadable and runnable on Windows. It is a fast neural network simulator with a graphical shell called 'Nutshell'. It is extendable with new neural network paradigms, though in practice this requires advanced knowledge of C++ and compilation techniques (and by now, access to some pretty retro compilation software).

In 2000, an unreleased Python version, walnut.py, was developed for research purposes in the NeuroMod group (see neuromod.org) and in 2004, Jaap Murre wrote a JavaScript version, mainly for for teaching purposes, which was used until recently: walnutJS 1.0 (see demos and lessons developed for a course on Connectionist Models), which was not officially released either.

Getting started

Installation

This project uses yarn workspaces so walnutjs-2.1 (npm package) and walnutjs-2.1-studio (vue3 frontend) can be easily installed in one go locally.

In the root folder of this git repository do:

yarn install

Running the tests

To run the test of the walnutsj-2.1 npm package (subfolder)

cd walnut-js
yarn test

Generate the docs

To generate the docs of the walnutsj-2.1 npm package (subfolder)

cd walnut-js
yarn docs

The docs will be found in walnutjs-2.1/docs.

Creating a basic network

import { Network } from "walnutjs-2.1/network/network";
import { Node } from "walnutjs-2.1/network/nodes/node";
import { Tract } from "walnutjs-2.1/network/tracts/tract";

import { BasicActivate } from "walnutjs-2.1/network/paradigms/examples/basicActivate";


// load a basic paradigm
const paradigm = new BasicActivate();

// create a network
const network = new Network("test", paradigm);
expect(network.name).toBe("test");

// add inNode
const inNode = new Node(network, "input", 4, 4);
network.nodes.addNode(inNode);

// add outNode
const outNode = new Node(network, "output", 4, 4);
network.nodes.addNode(outNode);

console.log("nodes so far:\n\n", JSON.stringify(network.nodes.dict, null, " "));

// add tract between inNode and outNode
const tract1 = new Tract(network, "tract1", inNode, outNode);
// connect sparse
tract1.connectBasicLinear(0.5, -1, 1);
// add
network.tracts.addTract(tract1);

// set some inNode neurons to activated
for(let i = 0; i < 8; i++){
  inNode.setNeuronAtIndex("net", i, 1.0);
}

// activate nodes
network.nodes.activate()

// check if activation activated some outNodes
console.log(network.nodes.neurons.net);

Basic example of a custom Paradigm

import { Paradigm } from "walnutjs-2.1/network/paradigms/paradigm";

class BasicActivate extends Paradigm {
 
  // activation rule ReLU
  nodeActivateFunction = (network, neuronIndex) => {
    const i = neuronIndex;
    const neurons = network.nodes.neurons;

    neurons.act[i] = neurons.net[i] > 0 ? neurons.net[i] : 0;
  };
  
  // basic activation function tracts,
  // net += act * weight 
  tractActivateFunction = (network, connectionIndex) => {
    const i = connectionIndex;
    const neurons = network.nodes.neurons;
    const conns = network.tracts.connections;

    neurons.net[conns.to[i]] +=  neurons.act[conns.from[i]] * conns.weight[i];
  };

  constructor() {
    super("basic activate");
  }
}

export { BasicActivate }

walnutjs-2.1-studio

Visual builder and analytics tool for walnutjs-2.1 networks.

Running locally

cd walnutjs-2.1-studio
yarn dev

more soon...