NeuralC

A simple c++ implementation of Neural Networks and optimization algorithms.

IMPORTANT!!!!

This is a program I made during summer 2017, it's not anything professional since it is slow, has close to zero comments and probably leaks tons of memory. Since that's the case, this repository has been archived and all previous commits messages deleted.

How to compile and run

In Linux:

1. Build with ./build.sh.
2. To run the program use ./NeuralC.
3. To install use ./build.sh -i
4. To uninstall use ./build.sh -u
5. To remove build files use ./build.sh -r

In windows: Use Linux :P

Console interface

neuralc [COMMANDS] [OPTIONS]

COMMANDS:
   SHORT FORM  LONG FORM   DESCRIPTION  
   -N          --new       creates a new neural network.
   -T          --train     Uses gradient descent to train a network.
   -I          --input     gets a vector from the user as input and prints the output.

OPTIONS:
    -n    --name: The name of the network (can be a path).
    -l    --layers: The number of neurons in each layer, separated by "/", i.e.: (2, 8, 1)=2/8/1.
    -d    --data: The name/path to the file with the data to train the network.
    -t    --threads: Number of threads to use.
    -i    --iterations: Number of iterations for the gradient descent.

EXAMPLE
    (1) Create the neural network: neuralc -N -n MyNet -l 2/8/1
    (2) Train it with the data in data.txt: neuralc -T -n MyNet -d data.txt -t 4 -i 1000000
    (3) Try the now trained network: neuralc -I 0.9/-0.45 -n MyNet

API Reference

Just include include/NeuralC.h in your project.

Files

• Network.cpp contains the Network class
• Optimization.cpp includes gradient descent

Copyright

Made by SERGIO MIGUÉNS IGLESIAS ([email protected]) for personal use, 2017.