Neural Net From Scratch in Javascript

Initialise a Neural Net, Forward & Back Propagate these inputs

The variables for each instance of input are in the columns of the input layer

The Truth contains the answers for each of those instances

There must be an input node for each variable

There must be an output node for each element in the structure of the truth

There are no hard and fast rules for the number of hidden layers

Somewhere in betweeen the number of inputs and outputs is not a bad place to start

The more there are, the easier it is for the neural net to come up with weights that work

But the number of calculations that need to be done rises

All the weights are seeded with random numbers that mean to zero. This means the process is stochastic and will vary from run to run

You can prefill the json input area with a number of examples

Choose some example json Binary converter 1 Binary converter 2 I Am Trask A sin wave Predictor with 5000 data points and biased input A sin wave Classifier with 5000 data points

{ "inputLayer": [ [0, 0, 0, 0, 1, 1, 1, 1], [0, 0, 1, 1, 0, 0, 1, 1], [0, 1, 0, 1, 0, 1, 0, 1] ], "truth": [ [0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7] ], "numberOfNodes": { "input": 3, "hidden": 4, "output": 1 }, "numberOfEpochs": 50000, "alpha": 1, "description": "The truth is a decimal representation of each binary digit divided by 10. This is a very simple type of problem to solve but is sufficient to test that the plumbing works. After 50,000 epochs, the mean of the absolute errors should be about 1%" }

Use the trained Neural Net and get it to predict the truth

The input field below has input as Json in the correct form for this network

You can choose a pretrained network or use the one trained from above

Use the one just trained An approximation of the sin curve between 0 and 2 pi