It's comes along with a matrix library to help with the matrix multiplications.
Right now the code is untested and only with basic checks, but I'm still working on it.
There's a s variable commented out in the code, it can be used to measure the error over iterations.
The error should get smaller as the MLP gets trained. The dataset used in the html example was taken from here.
Let's suppose you have the following data set:
|Height (cm)||Weight (kg)||Class (0-1)|
0 - adult
1 - child
You need to process the table to this format:
const x = [ [180, 80], [175, 67], [100, 30], [120, 32] ]; const y = [ [1,0], [1,0], [0,1], [0,1] ];
Then just create a new MLP passing the number of inputs, the number of nodes in the hidden layer, the number of outputs,
the learning rate and the number of iterations.
const nn = new MLP( x.length, x.length * 2, 2, 0.03, 500 );
Call the fit function
nn.fit( x, y );
And you're all set to make predictions
nn.predict( [178, 70] )
There's also a shuffle function that can be used before the training.
nn.shuffle( x, y );
You can take a look at the code here