To use it, you have to "See inside" the project. Currently the network is trained to find the XOR of two binary numbers. Here is a truth table for XOR: 1, 1 = 0 0, 1 = 1 1, 0 = 1 0, 0 = 0 What this means is that if you give 1, 0 as the input (as is the default in this project), you should expect to get a number close to 1 as the output (which you can see in this case is ~0.934). If you have experience with Brain.js, you can even use it to train your own model that you can use in this project! Assuming your neural network is called "net", you can use: "console.log(JSON.stringify(net.toJSON()));" to output your neural network as JSON, and then copy and paste it into the "json" variable inside the project.
This is a fully working and customizable neural network in Scratch. Features: - Compatible with Brain.JS - Should work with any sized network Known issues: - Only sigmoid activation function is implemented Credit: - TWY_ for their amazing JSON parser: https://scratch.mit.edu/projects/320131699/ Feel free to use this in your projects. Credit is appreciated!
