Gradient Free Deep Learning (GFDL) -- LANL O5013

Introduction

This is a Python library that provides a variety of scikit-learn conformant machine learning estimators that do not use backpropagation. The most prominent estimator types we support provide access to single and multi layer random vector functional link (RVFL) networks and extreme learning machines (ELMs). There is a considerable background literate on these two types of gradient free networks, and one obvious advantage of being gradient free is that expensive hardware devices are not required to train the models efficiently.

Contribution Guidelines

1. We use an open source license that is compatible with the rest of the scientific Python ecosystem, so please do not provide contributions that have a potential to be copyleft. For example, do not copy or even read code from libraries that have a GPL or other copyleft-style license, as we cannot accept it while retaining our more liberal software license.
2. At the moment it is not acceptable to use machine learning/AI/LLMs as part of the code contribution/review process. The reason is related to provenance and licensing---we cannot know for sure if the material being contributed originated or partially originated from code that had a copyleft license.
3. Please make an effort to format your PR titles and commit messages according to the guidelines used provided by NumPy. This helps keep our commit history readable and easier to debug.
4. Please try to avoid merging your own code---we aim to provide timely code reviews and have developers merge the code of others when they are satisfied.
5. Please add regression tests for bug fixes and new features, and avoid making unrelated changes (i.e., formatting changes to other parts of the code alongside a bug fix or improvement).