Micrograd C: Automatic Differentiation in C

Overview

Micrograd C is a lightweight implementation of automatic differentiation (autograd) in C, inspired by the Python micrograd library by Andrej Karpathy. This project provides a simple computational graph with backwpropagation support, allowing for gradient computation in basic mathematical operations.

Features

• Automatic gradient computation
• Support for basic operations:
  • Addition
  • Multiplication
  • ReLU activation
• Recursive backpropagation
• Memory management for computational graphs
• Interactive demonstration of autograd concepts

Prerequisites

• GCC or Clang compiler
• Standard C library
• Math library (-lm when compiling)

Compilation

Compile the program using:

gcc -o micrograd micrograd.c -lm

Usage

Run the program and follow the interactive prompts:

./micrograd

The program will guide you through:

1. Entering input values
2. Selecting a computation type
3. Displaying computational results
4. Showing gradients
5. Printing the computational graph

Computation Types

1. Multiplication + Addition + ReLU
2. Only Multiplication
3. Only Addition

Computational Graph

The implementation creates a computational graph that tracks:

• Node values
• Gradients
• Operation types
• Gradient requirements

Key Components

• Value struct: Core data structure for computational graph nodes
• Backpropagation methods for different operations
• Memory management functions
• Computational graph visualization

Example Output

Welcome to Micrograd C - An Autograd Demonstration
Enter first input value: 2
Enter second input value: 3
Enter bias value: 1

Select computation type:
1. Multiplication + Addition + ReLU
2. Only Multiplication
3. Only Addition
Enter your choice (1-3): 1

Computational Results
--------------------
Output: 7.00

Gradients
---------
Input1 gradient: 3.00
Input2 gradient: 2.00
Bias gradient: 1.00

Limitations

• Currently supports a limited set of operations
• Minimal error handling
• Designed for educational purposes

Acknowledgments

Inspired by the original micrograd implementation by Karpathy in Python.