What is pyNeuralNet ?

This project contain what you need to quickly set up, train and use a neural network. The model is trained on a supervised way. Once trained the network can be exported then reloaded during an other session, this allow you use the network on computer without any training.

Use

The NN2_0.py module contain a class named NeuralNetwork, that's all you need to import.

Creation

layers_shape = (2, 3, 3, 1)
NN = NeuralNetwork(layers_shapes)

This will create a neural network made of 2 input neurons, 2 hidden layers of 3 neurons each, and 1 output neuron.

Training

data = [np.array([[0],[1]]),
        np.array([[1],[0]]),
        np.array([[1],[1]]),
        np.array([[0],[0]])]

answers = [np.array([[1]]),
           np.array([[1]]),
           np.array([[0]]),
           np.array([[0]])]

NN.train(data, answers, 100000)

Here we want to train the network on a logical 'OR'. As the training is supervised, you must provide the correct answers to the model. Then you supply the number of training cycle you want, a cycle stand for a try->correction of the model.

Prediction

NN.predict(data[0])

To make a prediction, use the predict function with the input neurons value as parameter. This function will return a array containing values of the outputs neurons.

In case of multiple output neurons (ex: digit prediction) you can select the most activated neuron by passing the result of the predict function in numpy.argmax().

Import / Export

NN.load_network()
NN.export_network()

These functions will create or ask for weights.npy and biases.npy files in the current folder, they represent the model shape and values.