Caching and Reproducibility: Making Data Science Experiments Faster and FAIRer

Area of Science:

• Data Science
• Computational Science
• Software Engineering

Background:

• Research software in data science is often developed ad-hoc, lacking speed, reusability, and open access.
• This leads to significant researcher effort in reproducing or building upon existing work and risks experiment failure.
• Time pressure from iterative development and rerunning computationally expensive experiments is common.

Purpose of the Study:

• To propose integrating caching into the research software development process from the outset.
• To provide recommendations for developing research software that circumvents common issues like proprietary dependence and slow performance.
• To enhance the reproducibility and FAIRness of data within the open science workflow.

Main Methods:

• Outlining caching recommendations for research software development in data science projects.
• Demonstrating the utility of proposed caching strategies through a case study in mathematical information retrieval.
• Aligning recommendations with the Findability, Accessibility, Interoperability, and Reusability (FAIR) principles.

Main Results:

• Caching can significantly improve the speed and reusability of research software.
• Early integration of caching addresses challenges related to proprietary dependence and software failures.
• The proposed caching approach contributes to more reproducible scientific experiments and FAIRer data.

Conclusions:

• Making caching an integral part of research software development is crucial for data science.
• This approach enhances software speed, reusability, and reproducibility, aligning with open science and FAIR data principles.
• The recommendations presented are valuable for improving the efficiency and reliability of scientific research software.