Flexible_Neural_Net

A simple and flexible python library that allows you to build custom Neural Networks where you can easily tweak parameters to change how your network behaves

Installation

pip install flexible-neural-network

Initialization

• First initialize a Neural Net object and pass number of inputs, outputs, and hidden layers

  myNN = NeuralNet(number_of_inputs, number_of_outputs, number_of_hidden_layers)

• You can choose how what activation function to use from: "relu", "sigmoid, "tanh"

  myNN = NeuralNet(number_of_inputs, number_of_outputs, number_of_hidden_layers, activation_func="sigmoid")

• You can choose modify the learning rate

  myNN = NeuralNet(number_of_inputs, number_of_outputs, number_of_hidden_layers, learning_rate=0.1)

• You can choose tweak the number of nodes in each hidden layer

  • by assigning an integer number such as 3: if there was 4 hidden layers then each layer will have 3 nodes => [3, 3, 3, 3]

    myNN = NeuralNet(number_of_inputs, number_of_outputs, number_of_hidden_layers, nodes_in_each_layer=3)

  • by assigning a list of integers number such as [3, 5, 2, 3] that has a length of number_of_hidden_layers: if there was 4 hidden layers then each layer will have different number of nodes nodes correspondingly => [3, 5, 2, 3]

    myNN = NeuralNet(number_of_inputs, number_of_outputs, number_of_hidden_layers, nodes_in_each_layer=[3, 5, 2, 3])

How to use

Assuming you initialized your object and data as below:

myNN = NeuralNet(2, 1, 2, nodes_in_each_layer=4, learning_rate=0.1, activation_func="sigmoid")

data = np.array([
        [3,   1.5, 1],
        [2,   1,   0],
        [4,   1.5, 1],
        [3,   1,   0],
        [3.5, .5,  1],
        [2,   .5,  0],
        [5.5,  1,  1],
        [1,    1,  0]
        ])

mystery_data = [2, 1] # should be classified as 1

You can:

Here we specified the number of epochs to be 1

• Train single entries:
  myNN.train(data[0, 0:2], data[0, 2], epochs=1)

• Train multiple entries myNN.train_many(data[:, 0:2], data[:, 2], epochs=1)

• test single/multiple entries output = myNN.test(mystery_flower) where output is always an np.ndarray with size as the specfied in the object's constructor. for the current example it's = [1.45327823]

• Save NN for later myNN.save("file_name")

• Load NN without the need for retraining myNN = NeuralNet.load("file_name")

Obviously NNs do not give exact answers and its our job to determine which class is it belongs to and judging from the training data we only have class 0, or 1 and the output we got is nearer to 1 than 0 so we should classify it as 1