Creating Liquid Carrot

Table of Contents

• Part 1: Technology
  • Supporting Browser & Node.js
  • Testing Browser & Node.js
  • Threading (Node)
  • Threading (Browser)
  • Streaming (Node)
  • Streaming (Browser)
  • Documentation (Markdown)
  • Documentation (HTML)
• Part 2: Architecture
  • Connection
  • Neuron
  • Group
  • Network
  • Team
  • Population
  • Environment
  • Data
  • Bot

Part 1: Technology - Back to Top

Supporting Browser & Node.js - Back to Top

Articles

• Writing JavaScript Modules for Both Browser and Node.js
• Publishing packages that can be used in browsers and Node
• JavaScript module to run both in a browser and in Node.js
• Exports & Dynamic Dependencies for Node and the Browser

Examples

• Grunt, Browserify, Jasmine, Node Example - Dynamically Importing Modules
• Highland.js - Streaming in Node.js & The Browser - Dynamically Exporting Modules

Tools

• Browserify

Testing Browser & Node.js - Back to Top

TDD is best when implemented early - it saves lives, keeps users happy and secure, cures early-onset balding, reduces team-wide WTF's/minute, and makes your project look cool (i.e. build | passing tags).

Articles

• How to Write a Library for Node.js and Browser with Mocha Unit Tests
• UI Testing with Puppeteer and Mocha — Part 1 (Getting Started)
• Running Mocha Tests as Native ES6 Modules in a Browser
• Using Mocha with Puppeteer
• Inject jQuery into Puppeteer Page
• How to separate mocha tests in multiple files

Tools

• https://www.npmjs.com/package/mocha
• https://www.npmjs.com/package/chai
• https://www.npmjs.com/package/puppeteer
• https://www.npmjs.com/package/nodeunit

Threading (Node) - Back to Top

Articles

• Node.js cluster Documentation
• Node.js multithreading: What are Worker Threads and why do they matter?
• Multi threading and multiple process in Node.js
• Is it possible to achieve multithreading in nodejs?
• What are the drawbacks of making a multi-threaded JavaScript runtime implementation?
• Optimising* Node.js for Speed: Part 3

Tools

• https://www.npmjs.com/package/worker-farm

Threading (Browser) - Back to Top

Articles

• Run web worker with a function, rather than external file
• MDN Worker Documentation
• MDN navigator.hardwareConcurrency Documentation
• Get number of CPU cores in JavaScript?
• The Basics of Web Workers
• Naming a Web Worker for Future Reference

Examples

• http://jsfiddle.net/unKCK/
• JavaScript Web Workers in w/ One File

Tools

• https://www.npmjs.com/package/multithread.js

Streaming (Node) - Back to Top

Tools

• https://github.com/caolan/highland

Streaming (Browser) - Back to Top

Tools

• https://github.com/caolan/highland

Documentation (Markdown) - Back to Top

Tools

• https://www.npmjs.com/package/jsdoc-to-markdown

Documentation (HTML) - Back to Top

Articles

• How to JSDoc
• Auto-Generated Documentation Tools for JavaScript

Tools

• https://www.npmjs.com/package/toast-jsdoc

Part 2: Architecture - Back to Top

Serialization vs Parallelization vs Streaming vs Threading Table

Neural networks consist, primarily, of A LOT of neurons running dot product operations on data or results of "previous" neurons in the network and the squashing said dot product somehow - e.g. sigmoid function, ReLU function, etc. To run neural networks efficiently, we developers try to run operations in parallel using GPUs whenever possible. Here's a table outlining when the different dot and squash operations can be run in parallel (w/ GPUs).

Abstraction	Function	Serial	Parallel
Neuron	dot	true	true
Neuron	squash	true	false
Group	dot	true	true
Group	squash	true	true
Network	dot	true	true
Network	squash	true	true

Streaming vs ELT

Much like CPU vs GPU computing streaming and threading are complementary techniques that can further accelerate the speed at which a tool "runs". When using neural networks it's important that we squeeze out every computational resource that we can. Streaming allows us to consume less data and limit the amount of data that we consume per unit of time; threading allows us to process the data faster. Below we have a list of when we can use each technique.

Abstraction	Streaming	Threading
Neuron	false	false
Group	false	true
Network	true	true

Connections - Back to Top

Connections help track relationships between Neurons or Groups and move information among them.

Neurons - Back to Top

Neurons are the basic building block of neural networks - and cognition, in general.

Group - Back to Top

Groups are an abstraction to handle a collection of Neurons or other_Group_.

Network - Back to Top

Networks serve as the first unit of cognition and a collection of neurons and groups tied together.

Team - Back to Top

Teams serve as a collection of networks working together to solve a group of problems.

Population - Back to Top

Population is an abstraction to handle a collection of Networks or Teams.

Environment - Back to Top

Environment is standardized API that Populations, Teams, and Networks can consume to train or learn.

Data - Back to Top

Data is a utility API for Environments, Populations, Teams, and Networks to train or learn from data.

Bot - Back to Top

Bot is an abstraction of Data, Environment, Population, Teams, and Networks that focuses on non-technical ease-of-use.