Page Title

LobberFlow.js Documentation

Create an instance of the Neural_Network class with 2 hidden layers with relu activation, and an output layer with softmax activation:

            const model = new Neural_Network(structure=[
            {nodes: 16, input_size: 3, activation: "relu"}, //First hidden layer
            {nodes: 10, activation: "relu"}, //Second hidden layer
            {nodes: 2, activation: "softmax"} //Output layer
            ]);
            //First object of structure array must have input_size key
            //Supported activations are relu, sigmoid, and softmax(softmax only works for output layer)
        

Initialize the model with rprop optimizer:

model.setup(optimizer="rprop"); //Optimizer can be either "SGD" or "rprop"

Train the model on an extremely basic dataset over 100 iterations with an initial learning rate of .03 and no decay to the learning rate:

            var train_x=[[1, 0, 1], [0, 1, 0], [1, 1, 1], [0, 0, 0]];
            var train_y=[[1, 0], [0, 1], [1, 0], [0, 1]];
            model.run(train_x, train_y, epochs=100, learning_rate=.03, decay_rate=0);
            //Input data will probably be much more complex than this, and doesn't have to be 0s or 1s
        

Get a guess from the model:

            var x = [1, 0, 0];
            var guess = model.feedforward(x.concat(1)); //Must concatenate bias node, 1, to array
        

LobberFlow.js 1.0 Source and Documentation