Jordan "Jorb" Dehmel, 2023 github.com/jorbDehmel/jorbnet
JorbNet is a simple machine learning library for C++.
From this folder, call the command make install.
This library relies on SDL2 and clang, which must be
installed beforehand. Instructions on how to install
these based on your Linux distro are below. If you
are using Windows, install WSL/WSL2 with any of the
below distros.
Ubuntu / apt:
sudo apt-get install libsdl2-dev clang
Arch / pacman:
sudo pacman -S sdl2 clang
Fedora / yum:
sudo yum install SDL2-devel clang
After installation, you may delete this folder, although this may make updating and uninstalling jorbnet more difficult.
To update jorbnet, call the following command within this folder.
git pull && make install
If you no longer have this folder, call the following.
git clone github.com/jorbDehmel/jorbnet && make -C jorbnet && sudo rm -rf jorbnet
This will clone the source repo, install jorbnet, and delete the source code.
Simply call make uninstall from this folder, or call
the following command if you no longer have the
Makefile.
sudo rm -rf /usr/include/jorbnet /usr/bin/jorbnet-flags
Neither method will remove SDL2 or clang.
This is the base type of machine learning model. It is a Multi-Layered Perceptron (MLP) model which uses traditional backpropogation training.
This is a faster, but less efficient, way of training networks. It creates a pool of networks which are all trained at the same time using multithreading. These networks are periodically "pruned", leaving only the best. The best network is then cloned to repopulate the pool. It uses regular backpropogation training in addition to evolutionary pruning.
Activation functions are used in network propogation. Jorbnet includes several, namely the Sigmoid (as __sigmoid and __sigder) and ReLU (as __ReLU and __ReLUder) functions. To change an initialized network's activation function, set the following member variables. Note that you must change both for it to work properly.
Network n({1, 2, 3, 4}); n.act = __sigmoid; // Or __ReLU n.actder = __sigder; // Or __ReLUder
By default, networks use the Sigmoid activation function because it's my favorite.
For more information, as well as a mediocre derivation,
see the accompianing file docs/writeup.pdf (made via
make docs), or the sources listed below.
https://en.wikipedia.org/wiki/Backpropagation
http://neuralnetworksanddeeplearning.com/chap2.html
This library is protected by the GPLv3.