nikoladucak / neuralnet

Basic neural network implementation with command line interface.

Neural network implementation in C++

A simple feed forward neural net with stochastic gradient descent as learning method. Written in an effort to better understand innerworkings of neural networks. That adventure was made easier thanks to Neural Networks and Deep Learing e-book. Also,for efficient matrix multiplication Eigen linear algebra library was used.

Compiling

Dependancies: since Eigen is included within the source code, only boost in necessary.

Clone and navigate to root directory of the repo, then:

mkdir build
cd build
cmake ..
make

After that, nncli executable will be generated in the same directory.

Usage

Usage is centered around serialized neural network files. nncli provides commands to generate, train and feed input to serialized networks. Run nncli help for command line details.

Example

$ nncli test.nn make 2-3-1

Created test.nn with 2-3-1 m_topology.

$ nncli test.nn train example_dataset/xor.tre 1000 100 2.5

Starting training with:
        net: net.nn
        dataset: ../example_dataset/xor.tre
        epochs: 1000
        batch size: 100
        learning rate: 2.5
        training set size:4
        ....
Finished!

$ nncli test.nn feed 1-1
0.028531