Wekinator inspired a series of accessible Machine Learning tools for the web such as Teachable Machine and Teachable Machine 2. These teaching tools expand these hands-on supervised learning concepts into Physical Computing materials.
Version 2.0 builds upon the 2019 examples, streamlining the examples and adding further resources that compliment existing tools for Tensorflow Lite models. These templates provide students with a quick prototyping enviornment such as the P5JS Web Editor to generate graphical, audio , image outputs.
The code is open source and uses the ML5 Library and Neural Net Function
The repo for the examples lives here
Google’s Tiny Motion Trainer is a very powerful interface for training gestural data. However, it only captures the data (much like Teachable Machine) and if you want to do more with it you need to transfer the model data to other tools.
This example enables you to stream the trained data from Google's Tiny Motion Trainer Arduino example to P5JS
Steps:
This is an updated P5JS sketch that utilises the ML5JS Neural Net function. It allows you to send any number of inputs to P5JS and train them. This example no longer requires a template per different sensor; allowing changes to the settings variables to adapt to differring inputs.
Steps:
TRILLCRAFT
, ANALOG
, CUSTOM
let dataRange = [0, TRILLCRAFT];
const NUM_INPUTS = 24;
const LABLES = ["Square", "Circle", "Triangle"];
Once you have trained your model you may want to save it then pre-load it to work with when developing the outputs. The Physical Teachable Machine v2.0 pre-loaded model template helps you do this.