You ask me why?

For fun and pleasure. I’m testing few things and one part of it is solution for “black box” that will decide what to do based on incoming parameters. For example, my inputs will be:

I’m out of home
Person detected
Alarm enabled

Output:

send message to me

Of course I’m planning to heve few more input params.

So why not Neural network?

brain.js

Founded that lib, and looks very promising, but there is no out of box node to use in Node-Red. Anyway, thanks to a few sites and post, I have working flow:

How to install

First go to ~/.node-red dir, the one with lib, node_modules and settings.js and execute:

npm i brain.js

After installation edit settings.js. Look for functionGlobalContext and edit (as follows):

functionGlobalContext: {
  // os:require('os'),
  brainjs:require('brain.js')
}

After that - restart Node-Red.

Example

Example flow is trivial:

Function node

This is basic example and still doesn’t do what I want, but is perfectly fine as proof of concept.

Setup

I need to load brain.js lib

Function setup

On Start

Here I can create Neural network and train it. No need to do it on every message.

Function on start

On Message

And here is trivial part, parse parameters to network, and pass output to next node.

Function on message

Exported flow

[
  {
    "id": "4f7b8a1ffd9af2d2",
    "type": "tab",
    "label": "Flow 1",
    "disabled": false,
    "info": "",
    "env": []
  },
  {
    "id": "4403444239c81a4f",
    "type": "inject",
    "z": "4f7b8a1ffd9af2d2",
    "name": "",
    "props": [
      {
        "p": "payload"
      },
      {
        "p": "topic",
        "vt": "str"
      }
    ],
    "repeat": "",
    "crontab": "",
    "once": false,
    "onceDelay": 0.1,
    "topic": "",
    "payload": "",
    "payloadType": "date",
    "x": 120,
    "y": 80,
    "wires": [
      [
        "28b6e4a5c8517672"
      ]
    ]
  },
  {
    "id": "28b6e4a5c8517672",
    "type": "function",
    "z": "4f7b8a1ffd9af2d2",
    "name": "function 1",
    "func": "var net = context.get('net')\n\nmsg.payload = net.run([1, 0]); // [0.987]\n\nreturn msg",
    "outputs": 1,
    "timeout": 0,
    "noerr": 0,
    "initialize": "// Code added here will be run once\n// whenever the node is started.\n\nconst config = {\n    binaryThresh: 0.5,\n    hiddenLayers: [3], // array of ints for the sizes of the hidden layers in the network\n    activation: 'sigmoid', // supported activation types: ['sigmoid', 'relu', 'leaky-relu', 'tanh'],\n    leakyReluAlpha: 0.01, // supported for activation type 'leaky-relu'\n};\n\n\n// create a simple feed forward neural network with backpropagation\nconst net = new brain.NeuralNetwork(config);\n\nnet.train([\n    { input: [0, 0], output: [0] },\n    { input: [0, 1], output: [1] },\n    { input: [1, 0], output: [1] },\n    { input: [1, 1], output: [0] },\n]);\n\ncontext.set('net', net)",
    "finalize": "",
    "libs": [
      {
        "var": "brain",
        "module": "brain.js"
      }
    ],
    "x": 300,
    "y": 80,
    "wires": [
      [
        "3820ad2482c3582c"
      ]
    ]
  },
  {
    "id": "3820ad2482c3582c",
    "type": "debug",
    "z": "4f7b8a1ffd9af2d2",
    "name": "debug 1",
    "active": true,
    "tosidebar": true,
    "console": true,
    "tostatus": false,
    "complete": "true",
    "targetType": "full",
    "statusVal": "",
    "statusType": "auto",
    "x": 490,
    "y": 80,
    "wires": []
  }
]

Sources

Links to pages, that helped me:

Integrating brain.js with Node-Red