jamgrad documentation

jamgrad is a lightweight automatic differentiation library built from first principles.

Overview

jamgrad implements reverse-mode automatic differentiation (autograd) in Python and NumPy, forming the foundation for building and training neural networks with gradient descent.

Features

  • Computational graph construction and visualization

  • Automatic gradient computation via backpropagation

  • Support for common operations (add, multiply, power, exp, log, etc.)

  • DOT graph generation for visualization

Quick Start

from jamgrad import Tensor

# Create tensors with gradient tracking
x = Tensor([2.0], requires_grad=True)
y = x ** 2

# Compute gradients
y.backward()
print(x.grad)  # dy/dx = 2x = 4.0

Contents: