openGPMP welcomes any and all contributors! There are many ways to contribute to openGPMP. New implementations, updates to source code, documentations, language API, and much more. To submit a change simply fork the repository and begin the pull request process. Squash commits, clean up misc. files, etc. and submit a pull request to the openGPMP repository.
Be sure to include a hash # prefixing your commit messages. For example #EDITS:
or #UPDATES: are used to signaling commits that feature minor edits/updates.
#BUILD: references anything to do with the openGPMP project's build system.
#BUG: references a bug fix commit, and #FEAT represents a new feature being
added to the project. The prefixes, #PATCH, #MINOR, #MAJOR, are reserved
for automated versioning/tagging of commits based on messages. This process still
features a manual intervention by design meaning the maintainer of this project
will be in charge of publishing versions.
These are the current fixes, ideas, and implementations involved with openGPMP. Contributors are welcome to open a pull request, the maintainer (me), will take care of assuring passing tests and checks as far as github workflows.
-
At a glance openGPMP aims to feature fully functional Calculus, Linear Algebra, Number Theory, Machine Learning (and more) modules for modular and pre-built use.
-
The
Machine and Deep Learning Modulea continuous effort and I find is the most important and reusable module of the openGPMP project for programmers, students, researchers, etc. also meaning it requires the most help regarding implementations, tests, verifications, and any effort helping assure the accuracy of the implemented models.- Proper implementation of various styles of Nerual Networks, Regression methods,
classifiers, optimizers, accelerators, and more. So far I implement
the components of the Machine Learning module as I have time and as I familiarize
myself with a particular topic using some handy texts, papers, and other resources..
- The ML module will heavily rely off of the Linear Algebra module as well as
statistical methods that could perhaps be included in its own
statisticsmodule.
- The ML module will heavily rely off of the Linear Algebra module as well as
statistical methods that could perhaps be included in its own
- To prove the existing and future implemented models and methods provide some use, benchmarking/comparisons against industry-grade and open-source projects like PyTorch, TensorFlow (including the Keras interface), and Sci-kit Learn will prove beneficial.
- Proper implementation of various styles of Nerual Networks, Regression methods,
classifiers, optimizers, accelerators, and more. So far I implement
the components of the Machine Learning module as I have time and as I familiarize
myself with a particular topic using some handy texts, papers, and other resources..
-
Perfecting the Linux and OSX installations with
CMake.-
Windows is not currently on my agenda. Installation and support for Windows systems is handed off to those who need it or if I have a plethora of spare time.
-
CMakeinstallation currently only works with Linux, issues arise on OSX when it comes to the steps regarding GoogleTest.
-
-
Creating bindings for languages.
-
The stable
Python APImakes use of SWIG and distutils to wrap the source C++ code into a shared library, making use of Github Actions to publish the package to PyPI with support for just about allLinuxdistros andOSXversions and platform architectures, see here. If this process can be polished it is highly welcomed. -
Properly implementing, testing, and updating the unstable/inprogress APIs for
R,OCaml, andFortran.-
This will require knowledge of the languages and openGPMP in order to create samples for the respective languages in the samples folder.
-
As of now partial working steps for doing this with SWIG is present in each of the respective languages folders.
-
-
-
Formalize openGPMP for real world applications.
- Threading, acceleration, optimization, etc. This could include creating a 1-stop-shop implementation using some sort of cross-platform library/framework/technique that enable CPU and/or GPU exploitation just like popular libraries that are capable of handling extremely large data sets (PyTorch, TensorFlow, etc.).
- pthreads, openCL, openMP? A custom take at this could prove difficult to implement and maintain.
-
Updates to Documentation published with Doxygen to https://akielaries.github.io/openGPMP/. This is needed for both the C++ and Python uses while the former contains a decent amount already.
- If you have a bug fix please submit a pull request with details highlighting what you are fixing, how you replicated it, and how you tested your fix. For fixes that include confidential information please feel free to send me an email [email protected]
openGPMP attempts to follow the neat Google C++ Style Guide for naming while a custom clang-format file is used for tidying formatting.
Hopefully these guidelines are clear enough for you to begin contributing to this project. As always email me [email protected] for questions.