Micrograd-C

Micrograd-C is a simple automatic differentiation library implemented in C, basically a 'C' version of Karpathy's Micrograd. It provides a basic framework for creating and manipulating scalar values with gradient support, allowing for backpropagation through a computation graph.

Features

• Automatic Differentiation: Compute gradients for scalar operations using backpropagation.
• Scalar Operations: Supports addition, multiplication, power, ReLU, sigmoid, and tanh functions.
• Backward Pass: Propagate gradients through a computation graph.

Getting Started

To get started with Micrograd-C, clone this repository and compile the provided code.

Prerequisites

• A C compiler (e.g., gcc)
• A C++ compiler (e.g. g++)

Building the Project

1. Clone the repository:

   git clone https://github.com/shivendrra/micrograd.c.git
   cd micrograd-c

2. Compile the code:

• For c version

  gcc csrc/scalar.c test.c -o test -lm

• For cpp version

  g++ cppsrc/value.cpp cppsrc/module.cpp  main.cpp -o main -lm

Running the Example

After compiling, run the example:

• For c version

./test

• For cpp version

./main

Example Output

Here is an example of what you can expect when running the program:

Scalar[data=(2), grad=0.0074]
Scalar[data=(3), grad=0.0049]
Scalar[data=(5), grad=0]
Scalar[data=(6), grad=0.0025]
Scalar[data=(6), grad=0]
Scalar[data=(1), grad=0]
Scalar[data=(0.9975), grad=1]

Code Overview

scalar.h / value.h

Header file defining the Scalar structure and function prototypes for scalar operations and backward propagation.

scalar.c / value.cpp

C file containing all the necessary functions for Scalar value structure for ops & backprop.

nn.h / module.h

Header file defining the Neuron, Layer, & MLP structures & function prototypes for creating a small MLP.

nn.c / module.cpp

C file containing all the basic functions for building & implementing MLP in C using Scalar values.

test.c / cppsrc/main.cpp

Demonstrates the usage of the Micrograd-C library by creating scalar values, performing operations, and computing gradients.

Functions

• initialize_scalars: Initializes a scalar value with a given data value and its children.
• add_val, mul_val, pow_val, relu, sigmoid, tan_h, sub_val: Functions for scalar operations.
• backward: Computes gradients for all scalar values in the computation graph.
• print: Prints the scalar data and gradient values.

Contributing

Feel free to contribute by submitting issues or pull requests. Your contributions are welcome!

License

None!