openGPMP
Open Source Mathematics Package
Changelog

v1.0.0-rc.1

What's Changed

Reverted many changes from v0.10.0.

Documentation

Testing and Analysis

  • Updated some unit tests to increase overall coverage
  • Included static analysis workflows with clang-tidy, clang-analyze, cppcheck, osv-scanner, and infer

Core

  • Base DataTable, Logging, Threading, and other miscellaneous utilities

Disrete

  • Added class related to graphs

Linear Algebra

  • Mass rewrite to support AVX, SSE, and ARM platforms for now
  • Support for matrices as arrays and vectors
  • For now there are two main paths for this module. One is using somewhat naive methods that use naive blocking tricks and SIMD intrinsics, the other is the formal BLAS routine type of route that makes use of cache alignment, large blocking, assembly kernels, SIMD registers (AVX, SSE, MMX, ARM for now)
  • Added methods for eigenvalues, SVD, and linear systems, and (tensor methods which are still a large TODO)

Machine Learning

  • Added several naive implementations for neural networks, classifiers, regression methods, and more
  • These will need to be optimized and using the linalg module

Statistics

  • Added classes for probability distribution and descriptive statistics methods

tinygpmp

  • This is a subproject intended for MCUs (AVR and STM for now)
  • Big TODO as there is nothing formally implemented and just tests for now

Build System

  • Added options for building tinygpmp, including OpenBLAS in linking
  • Additional options for building tests, specifying tinygpmp build and microarch

ISSUES

Some known issues and things that need attention:

  • Majority of the project is not optimized and contains "naive" implementation with no regard for execution speed
  • Use of clang/clang++ for compilation produces some issues not seen with gcc/g++
  • Lots of duplicate code (everywhere for the most part, especially in the unit tests directory)
  • Test coverage is minimal and many methods are not tested as strongly as they should be
  • And some more...

v0.10.0

New contributions from @eeddgg, @sidsbrmnn, @igorsteinmacher

What's Changed

  • Project renamed from openMTPK to openGPMP. Python API renamed from openmtpk to pygpmp. Experimental Julia API as gpmp.jl
  • Documentation layout + additions
  • gpmp::accl namespace introduced hardware acceleration methods using GPUs with the help of OpenCL and CUDA for supported platforms (still very much experimental) and additional methods for CPU exploitation with Intel specific SIMD intrinsic functions. The acceleration unit is an optional install with openGPMP with the use of cmake.
    • gpmp::accl::GPU() class features methods of GPU acceleration using OpenCL. This class will house methods seen in the mathematical openGPMP modules that are ideal for GPU computation. The GPU functions themselves will be written in C with wrapper methods for use in C++ using extern "C" {/* function definitions */} making them available under the gpmp::accl namespace.
    • gpmp::accl::SIMD() class features accelerated functions seen in the core mathematical openGPMP modules ideal for wide register computations using Intel specific intrinsic functions.
  • gpmp::core::ThreadPool() class capable of spawning a specific number of threads or the default std::thread::hardware_concurrency().
    • view docs here > Note This class is not yet used innately within openGPMP and is provided as a user utility for now.
  • gpmp::core::ThreadDispatch() class that dispatches functions to the threadpool.
  • gpmp::core::DataTable() class capable of reading in CSV files storing them as a pair of vector strings. For now, the DataTable class features very limited functionality, reading in CSVs(entirely, or specific columns), formatted displaying, and two conversion methods for the string->int & string->double.
  • gpmp::core::TypeCast() class featuring several methods to convert C++ std:: specific data types to C-compatible data types. This class is utilized for the C++ wrapping of the C-based acceleration methods.
  • Created rolling branch that will update its commits from the gpmp-bot user on a nightly basis, using the #NIGHTLY: commit message prefix.
  • Introduced updates to calculus module in numerical differentation methods. First, second, and third order derivatives methods of a given polynomial in a function at a given value. For example f(x) = x^2ln(x) = x*x*log(x).
  • Updates to unit tests and additions for the new classes introduced.
  • Updates to namespaces, the core module containing the ThreadPool and DataTable related methods along with other miscellaneous tools lies under the gpmp::core namespace.
  • Patches introduced for #27 OSX Build reaches dead state
    • This introduced additional testing and CICD work related to openGPMP on OSX

v0.9.4

What's Changed

  • Division between code for internal functionality vs. what is available to the user.
  • CI/CD pipeline updates
    • clang-format for linting/formatting C++ source code, autopep8 for Python.
    • Python PyPI packaging edits.
    • Tag bump related commits will include a #MAJOR, #MINOR, or #PATCH prefix to the commit message.
    • Updates to certain directories run certain jobs.
  • Continuous Machine Learning & Linear Algebra module updates.
    • Stubbed out implementations for activation, regularization, optimization & model trainer methods, statistical helpers, random sorts + utilities.
    • Matrix related methods are usable. More updates to come.
    • ML module is not currently usable aside from Linear Regression model on very select data sets.

v0.7.0

New Changes in v0.7.0

  1. Updates to Primary and Secondary Multi-Layer Perceptron networks
  2. Updates to Python API
  3. Added support for Operating Systems, Kernels, Architectures, etc.
    • Added tested support beyond x86 processors
    • ARMV6, ARMV7, ARMV8, RISCV64, S390X, PPC64LE
    • Raspberry Pi 3 & 4 support
  4. Fortran, OCaml, Go and Julia demoted to experimental APIs, focus remains on Python and now R bindings

Next :

  1. Fix CMake bugs for OSX installation. Issues with Googletest hangup
  2. Update Machine Learning Module with more techniques
  3. Inject source code updates into APIs
  4. Create formal check/test process for Raspberry Pi hardware

v0.6.0

New Changes in v0.6.0

  1. Continued changes to Linear Algebra Module + samples
  2. Continued updates to Python API, injecting new C++ source functionalities into the setup.py file used in packaging process for PyPI
    • Install with pip install openmtpk==0.6.1
  3. Updates to the gpmp namespace. Wrapped more classes in the namespace, does not affect Python API
  4. Changes/updates to datatypes to allow for maximum precision while also avoiding potential overflow as well as for conciseness > Note Some int and char types will be updated later on to ensure nothing breaks
    • int--------—>int64_t
    • long long-->int64_t
    • unsigned-—>uint64_t
    • double----—>long double
    • float------—>long double
  5. Adding make uninstall option for the project's build process with CMake. See the install/uninstall directions here.

v0.5.3

What's Changed

  1. Updates to Linear Algebra Module and samples
    • See this example to view operations on matrices
    • More updates to samples as development continues
  2. PyPI updates
    • Adjusting SWIG process to accommodate for template based classes and methods
    • pygpmp 0.5.14 release here
  3. Updated to CMake installation of source C++ code

Next :

  1. Create first of many Neural Network Models and Machine Learning techniques for the ml module
    • Update the overarching ml.hpp header to make use of implemented models and techniques > Note So far the Linear Regression technique works on select two-column data
  2. Progress on CPP->R & OCaml language binding process

v0.5.2

What's Changed

Note Do not expect much until v1.0.0

New contribution from @earies

Major Updates to Python PyPI packaging process.

  • Builds wheels for OSX and Linux platforms (Windows will not be supported)
  • See all supported platforms.
    • Manually tested on Kali(latest), Ubuntu(latest) & macOS Ventura 13.1 (latest). Works for most OSX and Linux distro versions.
  • New versions must be specified in the setup.py file and will be published to https://pypi.org/project/pygpmp/ on pushes to MAIN.
  • Install with pip install pygpmp
  • Tools available:
    • Arithmetic based operations
    • Primes
    • RC4
    • Ciphers

C++ source is currently supported on Linux based platforms.

  • Bugs with googletest on OSX.

v0.4.0

What's Changed

  • Created Python pip package for Python API. Install with pip install pygpmp
    • Functionality is limited to Primes, RC4, Ciphers, and part of the Arithmetic module
    • Fixes to template classes/methods cpp->python conversion

v0.3.2

CMake installation make Googletest as a dependency.

  • openGPMP tests must pass in order to build and install the static library
  • TEST -> BUILD -> INSTALL

v0.3.1

Minor Release with updates to CMake installation process.

  • Headers are now installed in nested fashion with a parent-level header representing each module. For example:
    • Files in a module directory such as the ml/ directory (linreg.hpp, logreg.hpp, mlp_network.hpp, etc.) will make up that modules header file(ml.hpp).
  • Will include googletest workflow into CMake installation
    • Build library -> Test library -> Install library

v0.3

Formal documentation created with Doxygen deployed to github pages, see here.

  • Documentation updates within headers
  • Linear Regression model testing
  • Additional work and plans for Machine/Deep Learning module
    • Multi-Layer Perceptron, Linear Algebra Matrix, Scalar, and Vector manipulations + testing
    • Formal API installation processes
    • Automate API installations with Actions

As always refer to the project README.

v0.2

Formal installation of source C++ API with CMake. See installation steps in the project README.

  • Linux is supported.
  • OSX testing is in progress.
  • Working Python API, user installation mandatory.
  • Unstable Fortran, OCaml, and R bindings.

Future

  • Fixes + testing + formal installation of language bindings.
  • Neural Networks, Classifiers

https://pygpmp-docs.readthedocs.io/en/latest/index.html

v0.1

Initial introduction beta release of openGPMP

Open-source (intended) mathematics package written in C++ with a working Python API allowing for the use of algorithms and equations seen in many mathematical topics. See the README for more details

Modules

Module Features Status
Arithmetic Operations, GCD, etc. Working
Calculus Derivative at value Limited
Linear Algebra Matrix Operations Limited
Machine/Deep Learning Linear Regression Limited
Number Theory Primes, Cryptography/Encryption Working

Issues:

  • No formal installation method
  • Test coverage
  • OCaml & R bindings are unstable.
    • C++->OCaml class translation
Generated by doxygen 1.9.1