NNET

NNET is a C# class library for modeling and training neural networks using various layer types, activation functions, cost functions and optimizers.

Modeling

You can create a network using the Network class:

public Network(object inputSize, Layer[] layers)

Input size supports int for one dimensional inputs and Vector2Int for two dimensional ones. The layers array specifies the layers within the network.

Example network:

using NNET;

Network network = new Network(10, new Layer[]{
  new FullyConnected(20),
  new FullyConnected(20),
  new FullyConnected(2)
});

Training

You can train your network using the Network.Backpropagate function.

Example training loop:

Vector[,] data = new Vector[1000,2];
// Initialize your data
float LR = 0.5f;
for(int epoch = 0; epoch < 10; epoch++){
  Console.WriteLine("epoch " + epoch);
  for(int i = 0; i < 1000; i++){
    network.Backpropagate(data[i,0], data[i,1], LR);
    Console.WriteLine("sample " + (i+1) + "/1000:\ncost: " + network.cost);
  }
}

Layer types

FullyConnected

A regular fully connected perceptron layer.

Configurable variables:

• output size

Convolution

A convolution layer.

Configurable variables:

• kernel size
• kernel number
• stride
• padding

MaxPooling

A max pooling layer.

Configurable variables:

• pool size
• stride

Flatten layer

A layer which flattens a matrix. (Converts it into a vector)

Activation functions

• Relu
• Sigmoid
• Softmax
• Tanh

Cost functions

• Mean squared
• Mean absolute
• Cross entropy

Optimizers

• MiniBatch
• SGD (Stochastic gradient descent)
• Momentum SGD
• Adam