Making a prediction

While this network isn't trained yet, we already can try to make it predict the output of an input, by typing:

propagate(net, [1.0, 2.0, 1.0])


Since the network is initialized with random weights, the output in this case will be a 2-element vector containing gibberish.

Training the network

We can train our network using the dataset we specified earlier:

train!(net, inputs, outputs)


If you need to specify things like the learning rate, momentum rate, number of epochs and batch size, you should check the documentation of the function train!.