TinyMLjs

A set of hardware independent web-page, client-side (No Internet required) programming tools to help bring Machine Learning Proof of Concept's to more students, by using the power of the web with javascript-tensorflowJS and the simplicity of the C/C++ microcontroller serial print command combined together using Google's WebSerial that works on any desktop (or laptop) computer.

Note: WebSerial using a polyfill works on a few Android phones such as some of the Google Pixel phones.

Demos

Skip to basic instructions

Acceleration x,y,z time-series Demos: Now works for any sensor other than Vision and Sound

1. Best acceleration Nano33BleSense Does not have latest features.
2. Best acceleration NiclaVision-Acceleration and Time of Flight
3. Best acceleration RAK2270 Sticker Tracker

Note: This is also for other sensors! Always the best so far x,y,x acceleration page. CSV and Senses work and ML works, classification works, cell phone (Apple & Android) motion works



4. Next Acceleration Draft: for nicla-vision a49-next-acceleration.html - model view working, live edit not
5. Next Acceleration Draft: for nicla-vision a50-next-acceleration.html - adding live model editing WORKS!
6. Next Acceleration Last: for nicla-vision a51-next-acceleration.html - tidy it up and test
7. Next Acceleration Draft: for RAK2270 Sticker Tracker a52-next-acceleration.html- GOOD for RAK, but mouse motion issue
8. Next Acceleration blanked out mouse motion a53-next-acceleration.html
9. Next Acceleration for RAK2270 trying to fix mouse motion a54-next-acceleration.html Since it is good for the classroom, works but decimals
10. Next Acceleration a55-next-acceleration.html fixing mouse motion
11. Next Acceleration a56-next-acceleration.html fixing mouse motion
12. Next Acceleration a57 This is good!
13. Next Acceleration a58
14. Next Acceleration a59

Basic instructions

• This is a hardware independent proof of concept. Use a sensor or two and get a Machine Learning Keras model working with an actuator on any microcontroller board.
• Once you know your model works, slow down and record your data into a CSV file for uploading to a professional cloud Machine Learning service like edgeImpulse.com or Arduino Cloud or just stay client-side with tinyMLjs and build another model.
• If you are staying with tinyMLjs, upload the CSV files to a model designer, train the model and save your tensorflowJS saved model
• If you are still staying with tinyMLjs (wow you are great!) load the saved model onto a webpage with webSerial connection to your actuator and test how well your =model works in real world settings.
• Or use the conversion program either downloaded to online that converts the TensorflowJS saved model into tensorflowLite format and then use xxd -i to convert the binary file into a c-header file that can be installed onto your microcontroller using your IDE or the Arduino IDE.

Made by Jeremy Ellis. Twitter @rocksetta, or @rocksetta@masto.ai or GitHub profile at https://github.com/hpssjellis
This Github at https://github.com/hpssjellis/tinyMLjs Use tinyMLjs at your own risk!

Design By Max S. Github at https://github.com/maxbiz100

Started Jun 4th, 2023 Version 0.1.0-2