CooLearning: playground

📖 Context

This project is based on tensorflow/playground.

It aims to bring interactivity to the user interface with USB controllers by using WebMIDI.

It has been authored at CREATIS Laboratory (Lyon, France).

🚀 Use

🌐 Online version

Available here.

⚡ Offline version

• Go to releases
• Download the coolearning-playground-v{version}.zip asset
• Unzip the archive
• Open index.html with chrome

🔨 Development environment

Automatic publishing and versioning are enabled on the master branch and handled by semantic-release.

Commit messages are expected to be formatted following the Angular Conventions.

git clone https://github.com/coolearning/playground.git
cd playground
yarn
yarn dev
# navigate to http://localhost:3000

💫 Features

There are two main categories of devices you can attach.

🖱️ Selector

This device usually provides a grid of colored backlit pads.

Use the grid to display the inputs, neurons and output of the current neural network.

• Short press action:
  • enable/disable inputs and output links
  • select/unselect neurons
  • select/unselect layer
• Long press action:
  • enable/disable neurons

🎛️ Controller

This device usually provides faders, potentiometers and buttons.

A controller has 3 main modes:

• Default mode:
  • When nothing is selected, you can bind controls to parameters through the ** mappings UI**
• Selection mode:
  • When neurons are selected, you can change their:
    • Source weights
    • Source biases
    • Source meta (learning rate, activation, etc.)
• Layer mode:
  • When a layer is selected, you can change its:
    • Neurons biases
    • Neurons meta (learning rate, activation, etc.)

🧶 Constraints

• chrome only compatibility
• selector at least 64 pads
• controller at least 8 faders
• d3 version 3 dependency
• No automatic configuration for unknown devices, you must provide a configuration file.
  • Selector example
  • Controller example

📚 Supported Devices

Novation Launchpad X

Novation Launch Control XL