PIMS - CSC Seminar: Dhavide Aruliah

Reproducibility should be the bedrock on which our collective scientific research literature rests. Numerous "irreproducible results" serve as counterexamples showing us how far we are from achieving that ideal.  At the same time, scientific inquiry is becoming more data-intensive, so open science is growing in importance. A significant obstacle to reproducibility in computational science is the fact that many scientists are self-taught programmers; researchers learning computing on the fly as their careers progress can usually not expect to be as proficient as professional software developers in managing scientific software reproducibly.

In this talk, we'll highlight some recent technologies (notably Jupyter notebooks, the scientific Python software stack, and Anaconda) to enable reproducibility in scientific computing. We'll also discuss a set of scientific software development practices founded in software engineering research. When adopted, these practices can improve the reliability of our scientific software and our scientific productivity. The ultimate goal is to support the sharing, preservation, and reproducibility of scientific workflows (including data & software) with appropriate software tools and practices in place.