Nerve (Alpha)

Nerve is a programming language built around simple neural networks. The language is a code golfing language that instead of providing a set of instructions or commands to golf down code, you teach a neural network to complete the challenge. It takes a string of characters and returns a string of characters. This is all powered by TensorFlow.js.

Alpha

The language is currently under construction. I was building it from the ground up and writing my own forward/back propagation. Then I learned enough about neural networks to understand there is no way that I can get the same speed as TensorFlow. So, currently the testing page is provided to play around with the underlying components. Also, the prototype for the editor is available.

Nerve Verbose

Nerve Verbose is built out of these lower level components called expressions. The expressions combine together collecting information about the neural network to be created. This way everything is known at the very end and can be optimized down such that it is just the TensorFlow objects. Learn more...

The following is an example of a Nerve Verbose neural network that can learn to map a to A and A to a.

network(
    mapping("Aa"), // single character input that can either be 'a' or 'A'
    layers(),      // create single layer to map input to output
    mapping("Aa")  // single character output that can either be 'a' or 'A'
)

After teaching only once to map ['a', 'A'] to ['A', 'a'] we can end up with the following neural network that properly does the operation. The learning process utilized the meanSquaredError and sgd with learning rate of 0.001. Note that the first expression is utilizing the expression API to be less verbose. Once the code is ran, it is converted to its most verbose setting in order to ensure it is properly prepared to be converted to either Nerve Short or Nerve Golfed.

network(
    mapping("Aa"),
    layers(
        layer(
            neuron(
                0.9990000128746033,     // weight (neuron 0)
                1.0000009536743164,     // weight (neuron 0)
                -0.0010000000474974513  // bias   (neuron 0)
            ),
            neuron(
                1,                      // weight (neuron 1)
                0.9990000128746033,     // weight (neuron 1)
                -0.000999000039882958   // bias   (neuron 1)
            )
        )
    ),
    mapping("Aa")
)

There currently is a test page available to see it in action.

Here is the above network in its most verbose mode as well as the Nerve Short.

network(mapping(switchchar("Aa")),layers(layer.data(2,2)),mapping(switchchar("Aa"))).memory(string("w%BE%7F%3F%08%00%80%3F%00%00%80%3Fw%BE%7F%3Fo%12%83%BA%E1%F0%82%BA"))

n.t("Aa").l(2,2).t("Aa")._("w%BE%7F%3F%08%00%80%3F%00%00%80%3Fw%BE%7F%3Fo%12%83%BA%E1%F0%82%BA")