NeuralNet

Artificial Neural Net

Implementation of feed forward artificial neural net in c++(c++11) with back propagation algorithm for supervised learning.

HOW TO USE:

Just download header file NeuralNet.h and one of the libraries from lib directory:

• libann.a (static linking)
• libann.so (dynamic linking)

API:

• Make new Neural Net:
  • NeuralNet(const std::vector & topology);

/*makes new net with 2 input neurons,
                     4 hidden neurons,
                     6 hidden neurons(another hidden layer),
                     1 output neuron
*/
NeuralNet net({2, 4, 6, 1});

• Propagate inputs:
  • std::vector propagate(const std::vector & inputs);

/*propagate input {0.5, 0.5} through net and get output*/
std::vector<double> output = net.propagate({.5, .5});

• Training Neural Net:
  • void backProp(const std::vector & examples, double tol, double alpha);

/*vector of training examples*/
std::vector<TrainingExample> ex;

/*make new training example TrainingExample(std::vector<double> && inputs, std::vector<double> && outputs);*/
ex.push_back(TrainingExample({.5, .5}, {.5}));

/*train with accuracy of 0.000001 and speed of learning 1.0*/
net.backProp(ex, 0.000001, 1.0);

• Update weights of Neural Net:
  • void updateNNWeights(const std::vector & new_weights);

/*change weights of nn with vector of new_weights.
sizes must be same!
*/
net.updateNNWeights({0.5, 1.0, 0.65});

• Number of weights:
  • unsigned WeightsCount() const;

/*get number of weights in nn*/
unsigned n = net.WeightsCount();

• Serialization/Deserialization of Neural Net:
  • static void serialize(const std::string & file_path, const NeuralNet & net);
  • static NeuralNet deserialize(const std::string & file_path);

/*serialize nn 'net' in file Net.txt*/
NeuralNet::serialize("Net.txt", net);

/*deserialize nn from file Net.txt*/
NeuralNet net2 = NeuralNet::deserialize("Net.txt");