cmpad

View page source

Compare Algorithmic Differentiation Packages

Version of this Documentation

cmpad-2024.9.30

Summary

This package compares freely available AD tools. It compares both speed and how the user implements the computation of derivatives. The algorithms included in the comparisons are chosen so that they as easy to test for derivative correctness, are of a varied type, and can scale from very small to very large problems. Currently, only C++ and Python languages are included in the comparisons. It is easy to add a new package to the comparisons. It would be harder to add a new algorithm to the comparisons because one would have to test for correct derivatives. It would be even more difficult to add a new language, but the package structure is intended to handle this.

Release Notes

Compare AD Release Notes

Get Started

get_started

Git Repository

https://github.com/bradbell/cmpad

Example Speed Comparison

see xam_main.csv .

Getting Started

see get_started .

Gradient Source Code Comparison

C++

Calculate Gradient Using Adept

Calculate Gradient Using ADOL-C

Calculate Gradient Using autodiff

Calculate Gradient Using CoDiPack

Calculate Gradient Using CppAD Jit

Calculate Gradient Using CppAD

Calculate Gradient Using CppAD CodeGen

Calculate Gradient Using Sacado

Python

Calculate Gradient Using Autograd

Calculate Gradient Using cppad_py

Calculate Gradient Using JAX

Calculate Gradient Using PyTorch

Documentation

Name

Title

get_started

Get Started Using cmpad

get_package

Getting AD Packages

run_cmpad

The cmpad Main Program

algorithm

Algorithms Used to Compare Packages

cpp

C++ Algorithmic Differentiation

python

Python Algorithmic Differentiation

release_notes

Compare AD Release Notes