Deep-C

An implementation of deep learning algorithms in pure C programming language, without any third-party dependencies. All written from scratch. Suitable for embedded systems or low-level coding projects where TensorFlow and PyTorch cannot be used.

The library currently contains the following units:

• MLPC (Multilayer Perceptrons in C) - fully connected dense neural networks with SGD or Adam optimization.
• DDPGC (Deep Deterministic Policy Gradient in C) - deep Q-learning with continuous actions.

Examples:

• Learning the saddle function with MLPC.
• Swing up pendulum problem with DDPGC.

Building and running on Linux

No prerequisites needed. Just run make in the top folder. The following files will be created:

• ./lib/mlpc.a - the static MLPC library.
• ./lib/ddpgc.a - the static DDPGC library.
• ./bin/saddle - the saddle function executable.
• ./bin/pendulum - the pendulum swing up executable.

Building and running on Windows

Open the ./vs/deep-c.sln solution in Visual Studio and build/run the desired example.

Documentation

The provided saddle and pendulum examples should sufficiently demonstrate how to use the library. Additionally, Doxygen can be used to build the API documentation from the source code comments. Alternatively, you may examine the comments in the MLPC and DDPGC header files.

Acnowledgements

If you find this code useful in your project/publication, please add an acknowledgements to this page.