Skip to content

Latest commit

 

History

History
68 lines (56 loc) · 2.95 KB

mission.rst

File metadata and controls

68 lines (56 loc) · 2.95 KB
Raw

Aesara Mission Statement

Introduction

We, the Aesara Core Team ("we"), have created this mission statement to provide users and developers of the Aesara Project (the "Project") with a clear description of the Project's goals and guiding principles. The Project includes multiple software libraries including Aesara, AeHMC, AeMCMC, and AePPL, so any references to the Project apply to these libraries jointly.

Background

The overall goal of the Aesara Project is to provide a rich, Python-first toolset for the symbolic representation, manipulation, and compilation of mathematical expressions involving arrays.

More specifically, we want Aesara to facilitate the prototyping and development of higher-level reasoning and automation within mathematical modeling disciplines such as statistical modeling and machine learning. We also want Aesara to defer productively to more specialized external libraries for specialized and lower-level computations.

In general, three basic principles guide our work:

  • Modular design
  • Composition of different ecosystems
  • Separation of concerns

Project Goals

The higher-level goals of the Aesara Project are:

  • Produce code that is hackable in pure-Python. This includes Aesara’s graph elements and compilation process.
  • Reduce the development overhead between high-level prototyping and the construction of computable expressions.
  • Enable expressions represented in Aesara to be lowered to multiple compilation targets, and make lowering highly customizable.
  • Make computable representations of mathematics easier to construct, and facilitate connections with symbolic mathematics.
  • Facilitate the development of domain-specific compilers by domain experts who are comfortable programming in Python.
  • Provide tools to implement “meta”-level logic that includes a framework and base implementations for defining and executing arbitrary rewrites of Aesara expressions.
  • Offer the ability to choose between computational stability, speed, and any other custom metric during compilation.
  • Avoid unnecessarily restrictive choices that prevent reasonable uses of the Project that were not foreseen by the developers.
  • Use our work to demonstrate the utility of symbolic reasoning outside of its already established uses within the Python community.
  • Establish a diverse group of contributors who share similar interests and provide unique perspectives on how we can accomplish these goals.

The Project also commits to producing high-quality software by

  • prioritizing testing and its automation;
  • maintaining a well-defined and tested feature surface with a higher weight towards stability and the improvement of existing functionality over adding new features;
  • prioritizing design consistency, simplicity, and the ability to extend core functionality;
  • prioritizing work on core bugs, performance, and stability so that others can reliably build upon the Project.