autodiff-AsiaUnionCS107 0.0.7

An Automatic Differentiation Package

Final Project - autodiff Python Package

AC207 Systems Development for Computational Science, Fall 2020

Group Number 7

• Sehaj Chawla
• Xin Zeng
• Yang Xiang
• Sivananda Rajananda

Overview

Differentiation, the process of finding a derivative, is one of the most fundamental operations in mathematics. Computational techniques of calculating differentiations have broad applications in science and engineering, including numerical solution of ordinary differential equations, optimization and solution of linear systems. Besides, they also have many real-life applications, like edge detection in image processing and safety tests of cars. Symbolic Differentiation and Finite Difference are two ways to numerically compute derivatives. Symbolic Differentiation is precise, but it can lead to inefficient code and can be costly to evaluate. Finite Difference is quick and easy to implement, but it can introduce round-off errors. Automatic Differentiation handles both of these two problems. It achieves machine precision without costly evaluation, and therefore is widely used.

Our package, autodiff, provides an easy way to calculate derivatives of functions for both scalar and vector inputs. It implements both of forward and reverse mode of automatic differentiation, and also extends to root-finding with Newton's method.

We invite you to use autodiff package and help us improve it!

The link for full documentation is here.