NeuralNetwork

Back Propagation Neural Network (BPN) training.

• This code is a fork from Xavier Provençal's NeuralNetwork fork from Bobby Anguelov's NeuralNetwork.

The creation and training of this BPN uses the MNIST database of hand written digits.

Compilation

Compile under Gnu/Linux

Using cmake (recommended)

• Using Synaptic Manager, install the required compilation tools

  sudo apt update
  sudo apt install g++ make cmake
  

• Compilation

  mkdir build
  cd build
  cmake ..
  make
  

• Debug VS Release Mode

  Since training of neural networks requires a lot of computation, one might find useful to compile a debug and a release versions (the release version should run significantly faster):

  mkdir debug && cd debug && cmake -DCMAKE_BUILD_TYPE=Debug .. && make -j
  cd ..
  mkdir release && cd release && cmake -DCMAKE_BUILD_TYPE=Release .. && make -j
  

• Run

  From the build folder

  ./trainNN -h
  ./evalNN -h
  

Compile under Windows

See Visual Studio's guide for CMake projects

Visualization and testing of a BPN

Go on M. Provençal's website