PyreNet

About

Intended Use

In short, this is a C++ neural network static library developed as a simple, elegant, multi-purpose solution.

To be a bit more elaborate, this library offers a simple interface for home-cooked reinforcement based deep learning projects. It is optimized for running in a multi-threaded environment, seeking to offer performance and simple, essential, features without the complexity endured from larger-scale libraries. This library does not currently support back propagation.

Thread Safety

The library is designed to be fully thread-safe and can be used in multi-threaded environments.

Quick Start

Installation

For a pre-built library file, check out our latest release. No additional dependencies are required.

Example Usage

#include "PyreNet.h"

int main() {
  // Define middle and output layers
  std::vector<PyreNet::LayerDefinition> layerDefs;
  layerDefs.emplace_back(50, PyreNet::LayerDefinition::activationType::relu);  // Middle (50 nodes)
  layerDefs.emplace_back(50, PyreNet::LayerDefinition::activationType::sigmoid);  // Middle (50 nodes)
  layerDefs.emplace_back(5, PyreNet::LayerDefinition::activationType::relu);  // Output (5 nodes)

  // Initialize the network
  PyreNet::NeuralNet nn(5, layerDefs);  // Defines the network to have an input size of 5
  nn.mutate_gaussian(0, 1);  // Mutates network weights from a gaussian sample with mean 0, standard deviation 1

  // Run a prediction on an input vector
  std::vector<double> predictions = nn.predict(std::vector<double>{0, 1, 2, 3, 4});
}

Features

Activations

Activation Type	Identifier
ReLU	LayerDefinition::relu
Linear	LayerDefinition::linear
Hyperbolic Tangent	LayerDefinition::tanh
Sigmoid	LayerDefinition::sigmoid
Step	LayerDefinition::step

Serialization

For convenience, all networks can easily be serialized and deserialized.

PyreNet::NeuralNet nn(5, layerDefs)

ofstream ofs("output.txt");
ofs << nn;

ifstream ifs("output.txt");
ifs >> nn;

Mutations

Gaussian

mutate_gaussian(mean, std, OptionalInt(layerIndex));

Mutates the weights via a gaussian distribution.

If the layerIndex field is specified, only that layer will be mutated. Indexing starts from 0 at the first set of weights.

Uniform

mutate_uniform(lower_bound, upper_bound, OptionalInt(layerIndex));

Mutates the weights uniformly by a modifier in the range [lower_bound, upper_bound].

Contributing

Feel free to make a pull request if you have any useful features or bug fixes. For inquiries, contact harloff@umich.edu.

Authors

Hunter Harloff - Lead Developer - Poppro

License

This project is licensed under the MIT License

Poppro/PyreNet