# Empirical Study on the Software Engineering Practices in Open Source ML Package Repositories

**Authors:** Minke Xiu, Ellis E. Eghan, Zhen Ming,  Jiang, Bram Adams  
**arXiv:** [2012.01403](https://arxiv.org/abs/2012.01403)  
**Date:** 2020-12-02  
**Summary page:** https://www.summarizepaper.com/en/paper/2012.01403v2/

## Key points

- Recent advances in AI, particularly in ML, have led to practical applications like virtual personal assistants and autonomous cars.
- Development, training, and deployment of modern ML technologies require technical expertise and resources.
- Public ML package repositories have emerged as a way for practitioners and researchers to discover and reuse pre-trained ML models.
- This study analyzes two popular ML package repositories - TFHub and PyTorch Hub - comparing their features, policies, package organization, package manager functionalities, and usage contexts against established software package repositories.
- The study identifies unique software engineering practices and challenges associated with sharing ML packages.
- Limited empirical data is available on the current state and challenges of these repositories due to their recent emergence.
- These repositories play a crucial role in facilitating effective reuse of ML models.
- The findings provide valuable insights for data scientists, researchers, and software developers looking to utilize shared ML packages.

## Summary

Recent advances in Artificial Intelligence (AI), particularly in Machine Learning (ML), have led to the development of various practical applications such as virtual personal assistants and autonomous cars. However, the development, training and deployment of modern ML technologies like Deep Learning require significant technical expertise and resources. To address this challenge, public ML package repositories have emerged as a means for practitioners and researchers to discover and reuse pre-trained ML models. This paper presents an exploratory study that analyzes two popular ML package repositories - TFHub and PyTorch Hub - comparing their features, policies, package organization, package manager functionalities and usage contexts against well-established software package repositories like npm, PyPI and CRAN. By examining these repositories, the authors identify unique software engineering practices and challenges associated with sharing ML packages. The study reveals that while there is limited empirical data on their current state and challenges due to their recent emergence as a phenomenon; they play a crucial role in facilitating the effective reuse of ML models. The findings from this study provide valuable insights for data scientists, researchers and software developers who intend to utilize shared ML packages. This refined summary highlights the significance of these repositories in enabling effective reuse of ML models; emphasizing that they offer valuable insights for practitioners looking to leverage shared ML packages.