cmicrograd is C implementation of Andrej Karpathy's(Twitter) micrograd(GitHub repo).

This is my first interaction with neural networks. The original Python code is well explained in the video "The spelled-out intro to neural networks and backpropagation" and notebooks. Thanks, Andrej, I learned a lot!

More visual explanations of what is a neural network, gradient descent and how backpropagation works can be found in this series.

The training loop is free of malloc() and free() for speed. The 'topo_mt.h' executes NN in multiple threads.

• All tests passed
• under 1K LOC
• Linux only

• Windows OS support
• read/write parameters on disk
• add example - https://github.com/karpathy/micrograd/blob/master/demo.ipynb
• more examples(recognizing handwritten numbers/letters, etc.)

git clone https://github.com/milansuk/cmicrograd
cd cmicrograd/linux
sh build_r
./cmicrograd_r

• /src
  • main.c - runs examples
  • examples.h
  • value.h - Value is node in neural netowrk
  • mlp.h - MLP neural network
  • topo.h - orders Values for execution(training)
  • topo_mt.h - executes Values in multiple threads
  • std.h - bridge to operation systems
• /linux - compile/run/debug scripts for Linux OS

Milan Suk

Email: [email protected]

Twitter: https://twitter.com/milansuk/

Your feedback and code are welcome!

For bug reports, please use GitHub's issues.

cmicrograd is licensed under Apache v2.0 license.