A general-purpose, lightweight feed-forward neural network library for C++
)
)
)
)
- Cross-platform compilation and targeting (x86/ARM)
- Customizable input, hidden, and output layer sizes
- Serializing/deserializing trained models, saving/loading training checkpoints, etc.
For the Pong demo, just compile using make and run using fnn.exe or ./fnn. This will show an example of this FNN library playing Pong and using reinforcement learning to improve over time. Almost always, main.cpp can be modified to fit your use case, and it is a great example to build from.
Note
Rendering is a bit strange depending on the platform. It renders fine on most Linux terminals, but both cmd.exe and Visual Studio Code's terminal fail to render the example correctly. On Windows, the Visual Studio 2022 Debugger Console renders the example correctly. On Windows, the example can also be compiled and ran using MSYS2. Performance also changes drastically depending on compiler settings.