This is my own Javascript implementation of a backpropogation neural network which uses a binary sigmoid (range 0,1) as the squash function. The data used to test the system is the well-known Iris data set.
Development environment = Visual Studio Code, Google Web Server for Chrome, and Chrome web browser. Not necessary for running the application.
Iris test dataset
The app has a single html page, no web service, and a single javascript file (main.js). It utilizes jquery which is easy to remove. All the math and code is plain vanilla js except for a few jquery calls.
The app starts up, immediately reads in the values from inputdata.js, targetdata.js, and testdata.js. The full Iris data file in the datastore_input.js with corresponding targets are in datastore_target.js.
You can play around with the settings like the number of epochs, learning rate, etc. All the settings are at the top of main.js.
Contact me if you have questions.
Enjoy!