Enabling Python Machine Learning Libraries on Windows ARM: Challenges and Solutions

Citation: Gleb Khmyznikov, "Enabling Python Machine Learning Libraries on Windows ARM: Challenges and Solutions", Universal Library of Engineering Technology, Volume 01, Issue 02.

Copyright: This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract

The article examines the emergence of Windows ARM as a machine learning platform through the lens of compatibility, performance, and the maturity of the Python ecosystem. The relevance of the study is determined by the spread of energy-efficient ARM devices running Windows and by the need for native execution of ML workloads without the losses introduced by x64 emulation. The aim of the work is to systematize the barriers to porting Python machine learning libraries to Windows ARM and to identify practical paths toward building a full-fledged development environment. The scientific novelty of the article lies in an integrated analysis of the compatibility of the scientific Python stack, architectural constraints, and adaptation mechanisms, including LLVM/Flang, ARM64 EC, DirectML, and the QNN Execution Provider. The work includes an experiment involving automated testing of Python packages in native ARM64 mode, x64 emulation, and a control x64 environment, as well as comparative performance benchmarking. It was established that native Python ARM64 execution provides a marked speed increase, whereas the main barriers remain SciPy’s Fortran dependency, limited support for a number of key ML libraries, and the immaturity of the build infrastructure. The article will be useful for ML engineers, Python library developers, and software platform architects.

Keywords: Windows ARM, Python, Machine Learning, ARM64, Scientific Libraries, Benchmarking.

https://doi.org/10.70315/uloap.ulete.2024.0102015