Welcome to Undefined - an automatic differentiation tool’s documentation!

Features

Be awesome and fast automatic differentiation calculation!!!

• We utilized the dual number approach in the forward mode to achieve fast computation.

• In the reverse mode, we implemented a two-step approach to recover the partial derivative, and eventually trace backwards for the final derivative results. For details about the implementation, refer to our implementation section.

• Our design of the software is friendly for the users in command-line interaction, and our codes are well encapsulated and modularized.

Implementation of the forward and reverse mode (extended feature) to automatically calculate derivative.

In the extended feature, our package can build computational graphs in the reverse mode.

Link to the project on Github page.

Link to the package on PyPI.

Introduction and Background

• About
  • 1. Introduction
  • 2. Background
    • 2.1 Chain Rule Formula
    • 2.2 Elementary Function
    • 2.3 Computational Graph
    • 2.4 Automatic Differentiation

Getting started

• Usage
  • 3. How to use Undefined
  • 3.1 Forward Mode Demo
  • 3.2 Reverse Mode Demo (Our Extension)
  • 3.3 Multiple Vectors Inputs and Outputs
  • 3.4 Seeds Vector option
  • 3.5 A Few Tips
  • 3.6 Debugging

Software Organization

• Testing and Software Organization
  • 4.1 Test Undefined
  • 4.2 The directory structure looks like the following

Implementation

• Implementation
  • 5.1 Core Data Structure
    • Forward Mode
  • 5.2 Extension Implementation
    • Reverse Mode
  • 5.3 Core Classes
  • 5.4 External Dependencies

Broader Impact and Future Directions

• Broader Impact and Future Directions
  • Broader Impact and Inclusively Statement
  • Future Directions
    • Beyond the first order derivative
    • Graphic User Interface (GUI)
    • Integrated with ML models

Code details

• undefined
  • undefined package
    • Submodules
    • undefined.API module
    • undefined.Calculator module
    • undefined.GraphGenerator module
    • undefined.UDFunction module
    • undefined.Utils module
    • Module contents

Licensing

We will use the MIT license for open source software development so that other people who are interested in our software will have access to contribute.

Reason for our choice: We want it to be simple and permissive. Under the MIT license, anyone can contribute to this project by adding functionality, debug, or customize it to meet their needs. Please see the LICENSE file in your github for details.