My most recent paper submission (preprint available) is about improving the verifiability of computer-aided research, and contains many references to the related subject of reproducibility. A reviewer asked the same question about all these references: isn’t this the same as for experiments done with lab equipment? Is software worse? I think the answers are of general interest, so here they are.
A recent article in “The Atlantic” has been the subject of many comments in my Twittersphere. It’s about scientific communication in the age of computer-aided research, which requires communicating computations (i.e. code, data, and results) in addition to the traditional narrative of a paper. The article focuses on computational notebooks, a technology introduced in the late 1980s by Mathematica but which has become accessible to most researchers only since Project Jupyter (formerly known as the IPython notebook) started to offer an open-source implementation supporting a wide range of programming languages. The gist of the article is that today’s practice of publishing science in PDF files is obsolete, and that notebooks are the future.
It is widely recognized by now that software is an important ingredient to modern scientific research. If we want to check that published results are valid, and if we want to build on our colleagues’ published work, we must have access to the software and data that were used in the computations. The latest high-impact statement along these lines is a Nature editorial that argues that with any manuscript submission, authors should also submit the data and the software for review. I am all for that, and I hope that more journals will follow.
Data science is usually considered a very recent invention, made possible by electronic computing and communication technologies. Some consider it the fourth paradigm of science, suggesting that it came after three other paradigms, though the whole idea of distinct paradigms remains controversial. What I want to point out in this post is that the principles of data science are much older than most of today’s practitioners imagine. Let me introduce you to Apollonius, Hipparchus, and Ptolemy, who applied these principles about 2000 years ago.
The plea for stability in the SciPy ecosystem that I posted last week on this blog has generated a lot of feedback, both as comments and in a lengthy Twitter thread. For the benefit of people discovering it late, here is a summary of the main arguments and my reply to them.
Two NumPy-related news items appeared on my Twitter feed yesterday, just a few days after I had accidentally started a somewhat heated debate myself concerning the poor reproducibility of Python-based computer-aided research. The first was the announcement of a plan for dropping support for Python 2. The second was a pointer to a recent presentation by Nathaniel Smith entitled “Inside NumPy” and dealing mainly with the NumPy team’s plans for the near future. Lots of material to think about… and comment on.
It’s hard to find an aspect of modern life that is not influenced in some way by software. Some of it is very visible, for example the Web browser I start on my computer. Other software is completely invisible, such as the software controlling my car’s diesel engine. Some software is safety critical, for example flight control software in airplanes. Other software is used in a much more futile way, such as playing games. I could go on listing characteristics in which different software packages differ, but I will leave it at that - I don’t really expect anyone to disagree about the ubiquity and diversity of software in our increasingly digital world.
A few days ago, I noticed this tweet in my timeline:
I 'still' program in C. Why? Hint: it's not about performance. I wrote an essay to elaborate... appearing at Onward! https://t.co/pzxjfvUs5B— Stephen Kell (@stephenrkell) September 5, 2017
That sounded like a good read for the weekend, which it was. The main argument the author makes is that C remains unsurpassed as a system integration language, because it permits interfacing with “alien” code, i.e. code written independently and perhaps even in different languages, down to assembly. In fact, C is one of the few programming languages that lets you deal with whatever data at the byte level. Most more “modern” languages prohibit such interfacing in the name of safety - the only memory you can access is memory allocated through your safe language’s runtime system. As a consequence, you are stuck in the closed universe of your language.
Over the last few years, I have repeated a little experiment: Have two scientists, or two teams of scientists, write code for the same task, described in plain English as it would appear in a paper, and then compare the results produced by the two programs. Each person/team was asked to do a maximum amount of verification and testing before comparing to the other person’s/team’s work.
A few years ago, I decided to adopt the practices of reproducible research as far as possible within the technical and social constraints I have to live with. So how reproducible is my published code over time?