Catherine (Katie) Gosmeyer recently joined our editorial staff as an Assistant Editor. She’s a Scientific Software Developer at NASA Goddard Space Flight Center and works on the ICESat-2 calibration team; formerly she was an analyst at Space Telescope Science Institute (STScI) working on Hubble and James Webb Space Telescopes. Katie is very interested in open source software and improving research and approached me after a presentation I gave at GSFC about volunteering for ASCL as an editor. Yes!! She has dived right in, and joins Associate Editor Kimberly DuPrie, who is at STScI, and me in creating and editing ASCL entries. Welcome, Katie!
BLOCK 6: Software Publishing Special Interest Group Meeting
This meeting-within-a-meeting was an opportunity for journal editors, publishers, referees, abstract services, and others associated with research software publication to discuss how best to include research software in the scholarly record, improve the sustainability and reproducibility of research articles, and share information on issues and possible solutions. Representatives from Science, Nature-Springer, MNRAS, Oxford University Press, and AAS Journals were among the journals and publishers attending. As the session was open to all, researchers and software authors also attended. The agenda had three main items on it: journal software policies, ratings for numerical reproducibility, and improving instructions for authors and referees. The session was moderated by Rein Warmels (ESO, DE) and Alice Allen (ASCL, US).
After introductions by all in the room, the first agenda item, journal policies on software, was opened for discussion. Keith Smith, associate editor for astronomy and planetary science for Science, shared that his journal is requiring that software that enables research results be shared. Editors from other journals stated that this would probably not work for them, though they are sympathetic to the importance of research software transparency. Chris Lintott, lead editor for Instrumentation, Software, Laboratory Astrophysics, and Data for AAS Journals, said that he rewrote the Journals’ software policies on his first day at the job to require formal citation of software. Smith pointed out the expectation is different for data; people are much more open about sharing observational data, and A&A (may) require it. Warmels noted a difference between data and software; the quality of observational data from an observatory is known. This is not the case with software. We cannot know the quality of the unreleased code. Amruta Jaodand asked whether publishing houses have software reviewers. The various astronomical societies’ journals do not peer-review code; there are a few journals that do perform code review of various depths, such as the Journal of Open Source Software and Software X, both of which focus on research software across disciplines.
There was support for better software citation, but not for ratings of articles for numerical reproducibility. The idea of ratings for reproducibility led to a discussion about reproducibility itself and the issue of releasing software written for research. Adam Leary, senior publisher at Oxford University Press, said that the Journal of Biostatistics rates the reproducibility of its articles and that that journal has a reproducibility editor. The group discussed the workload this might put on reviewers along with other issues, which would slow down the review. But the real job would be for the authors! Brigitta Sipocz mentioned the need for a feedback loop, which triggered the question as to why one would want to put a lot of resources into reproducibility. Smith replied that there are cases in other sciences where whole bodies of work could not be reproduced! Warmels pointed out that a number of fake results were found by the community, not by reviewers.
Someone suggested pushing the community toward releasing software through funding councils. Lintott initially liked the idea, and stated that journals could enforce this by checking papers against funding/funders that require release. Allen found this is an intriguing suggestion. Additional discussion raised several issues with this approach. The impracticality of implementing the idea became obvious when considering the time and resources it would take and the complexity of funding, as well as varied requirements of a large number of funding organizations.
The discussion then turned to reasons researchers do not release their software, with one advantage stated as, “If you don’t give me this funding, this research will not be done.” We have to change the way we argue for funding, then… “because as the only person who can do this, keeping my code private IS an advantage.” Jaodand mentioned that researchers get less credit for software than for research results. Smith replied that until we build up a virtuous cycle of code release, something he had mentioned in his presentation the previous day, the answer may be getting the credit system working first.
Another disincentive for code release mentioned was the possibility of someone running software incorrectly and then publishing “this code doesn’t work.” Lintott said that we should look for this and get data on it, so we can answer the question, “How often has this happened?” He also suggested looking for the positive cases, where release has been good for a developer or developer team, and provide this data to code authors.
The next agenda item was improving instructions for authors and referees on software citation and treatment. According to Smith, Science’s instructions were improved by rewiting them to accord with the Center for Open Science‘s Transparency and Openness Promotion (TOP) Guidelines. The guidelines are very helpful, and using them provides clear instructions. Greg Schwartz, data editor for AAS Journals, asked how we could better encourage authors to read instructions. Smith waggishly replied these instructions exist so journal editors can point to them. It was suggested that journals standardize their instructions not only to help authors out but also to discourage what was referred to as “research tourism.” The TOP Guidelines were again brought up as a good tool to use for standardizing instructions. Someone asked about having a section for acknowledgements or statements for software. Smith pointed out the danger that some may think that section as a substitute for formal citation. Allen agreed that software should have formal citations, and also stated her appreciation for the Software section that AAS Journals have added to their papers. The ASCL has long been interested in seeing such a software listing in research articles (in addition to, not as a substitute for, formal citation of software). Warmels returned the discussion to the idea of standardization of instructions, asking whether this can be done. Journal representatives said the various journals do get together to standardize where they can, and are due to do so again.
The discussion migrated to openness in general. Among the suggestions for moving the discipline to be more open were to “Advertise your openness!” and to include a slide in your presentations that say your work is open and reproducible; this lets your peers know that you value openness, and can help others think about working more openly. The point was made to not rely on policing for open practices as resources aren’t available to do so. The role of education was brought up, too: Researchers need to be taught how to make their data and software open.
The final agenda item was to decide whether an on-going software publishing special interest group might be welcomed by those in the room; there was no support for this. Journals already have a method to share information amongst themselves and everyone is oversubscribed to meetings, groups, and conference calls. With that item settled, the meeting and Software in Astronomy Symposium concluded.
The ASCL thanks the Heidelberg Institute for Theoretical Studies for its generous ongoing support, which permitted two participants in this symposium to attend the EWASS/NAM meeting who would not have been able to do so without it.
BLOCK 5: Machine Learning & Data Mining
Stephen Serjeant (Open University, UK) moderated the fifth session. This session presented different techniques to, for example, study noise in gravitational wave interferometers, select young stellar object candidates, and directly image exoplanets. David Cornu’s (UTINAM, FR) talk, titled Selection of Spitzer YSO candidates using deep learning classifier, included a short tutorial for creating an artificial neural network, showing how a small neuron takes input vectors and updates weights associated with them to understand anti- and co-relation between various factors. Similarly, Carlos Alberto Gomez Gonzalez (U Grenoble Alpes) showed how supervised machine learning can be used to detect exoplanets in his presentation Data science for direct imaging of exoplanets. Other talks in this session included Massimiliano Razzano (INFN, IT), Deep learning to study the noise in gravitational wave interferometers, Sebastian Turner (LJMU, UK) presenting on k-means clustering in galaxy feature data, and Emille Ishida (COIN, FR) presenting for Santiago Gonzalez Gaitan, with Spatial inference of astronomical datasets with INLA. Robert Lyon (UManchester, UK) finished this session with a presentation on Imbalanced learning in astronomy, and provides a Jupyter notebook containing a tutorial and examples. The presentations in this session were accessible even to those with no experience in data mining and machine learning, as the techniques used were explained quickly and well before moving on to how they enabled particular research.
Slides from this session
Astronomy leads most sciences in providing many open services, particularly data and ways to get access to and use data. This session, moderated by Andy Pollock (USheffield, UK), highlighted some of the new and ongoing services available to not just professional astronomers, but also to students and other interested parties. Debbie Baines (ESA, ES) opened the session with her presentation ESASky version 2: the next generation, which included a live demonstration of this incredible resource. ESASky allows searching for any astronomical object and offers viewing it in different wavelengths. It is very fast, too; Baines loaded images from Herschel, XMM, Chandra, and HST in her live demo. It offers relevant links to SIMBAD, such as to papers, and one can pull up research literature from ADS in ESASky, too. Rachael Ainsworth (UMelbourne, AU), originally scheduled to give the first presentation, followed Baines; her presentation, Open Science in Astronomy, gave an overview of openness in astronomy. She pointed out that astronomy is better at open science than many other fields, crediting, among other services, arXiv and GalaxyZoo. She covered some of the challenges and shared additional resources that support, encourage, or make open science possible. Jorge Palacios (IFAE, ES), in his talk Astronomy in a Big Data platform, discussed two services, Cosmohub, a web portal for interactive exploration and distribution of massive cosmological data, and SciPIC (Scientific pipeline at PIC), software for generating synthetic galaxy catalogs using DM simulations. In An Interactive Sky Map based on the Byurakan Plate Archive, Gor Mikayelyan (BAO, AM) shared that the NAS RA Byurakan Astrophysical Observatory in Armenia is making its plate archive, covering 1947 -1991, available online in the BAO Observational Database, which will have the ability to search and select and will allow downloading the plates in different formats. A sky map will show where plates are available. After the last two talks, The ASI Cosmic Ray Database for charged particles data by Valeria Di Felice (SSDC/INFN, IT) and Using XML and semantic technologies in astroinformatics to manage data by Guy Beech (UHuddersfield, UK), Pollock opened the floor for discussion and asked whether people could do what they’d like to do as efficiently as they’d like. One answer from the audience was no, because the data are heterogeneous and different ways are required to access them. One pointed observation arising from the back-and-forth was that “radio astronomy is decades behind” in terms of software services, with “thousands of different file formats processed by thousands of different programs”.
Slides from this session
BLOCK 3: Software packages for research
Amruta Jaodand (ASTRON, NL) moderated the third session of the Symposium, which featured talks on seven different software packages. The first presentation was by Sergio Martin (ESO, CL), on MADCUBA and SLIM: A lightweight software package for datacube handling and spectral line analysis. These software packages offer features for easier use of the ImageJ image processing framework; this framework is widely used in other disciplines for multidimentional images. SLIM provides a synthetic spectra generation and automatic fit to observed spectra; MADCUBA provides an interface to handle and manipulate multiple datacubes, process large datasets quickly, and is scriptable. Alex Hamilton (UHull, UK) presented SunPy the Open Source Solar Physics Library, giving an introduction to the package, which collects, manages, and analyzes data from many solar data sources, and sharing information on its new features and capabilities.
As one of this session’s presenters was doubled-booked, we made adjustments to the order of presentations, so The next-generation cosmological code SWIFT was the next talk; this was given by Matthieu Schaller (LeidenU, NL). This package, due for release this summer, uses task-based parallelism for intra-node parallelization; testing has demonstrated that it is more than 30x faster than the code Gadget on representative cosmological problems while using fewer resources. The next talk had generated a lot of interest, as it is not a usual astronomy conference offering. Maisie Rashman (LJMU, UK) spoke about Developing and applying astronomical software for novel use in conservation biology. The team Rashman works on has developed a pipeline using astronomy techniques to identify and track animals; their goal is to create a fully automated system for species identification, population tracking, and combating poaching using drones. After Rashman, Shane Maloney (Trinity College, IE) presented xrayvision – a collection of image reconstructions methods for X-ray visibility observations. The xrayvision package is built atop other packages, including SunPy, and fills a need for an open-source solution for these observations. One of the project’s goals is to provide access to people in poorer countries to software for solar physics. The cleverly-named software package PampelMuse was presented by its author Sebastian Kamann (LJMU, UK) in his talk Crowded-field 3D spectroscopy with PampelMuse. This Python software performs PSF deblending on the integral-field data; several thousand sources can be deblended simultaneously. The final talk of this packed and fast-paced session was by Matteo Bachetti (INAF, IT) on Stingray, HENDRICS and Dave: Spectral Timing for all. Stingray is an AstroPy affiliated package that merges timing and spectral analysis for X-ray spectral timing. The idea for spectral timing arose in a meeting at Leiden University two years ago, where a team formed and started working on the project; the code is under active development. By combining Stingray with the HENDRICS (for shell scripting) and Dave codes (a GUI atop Stingray), the team will provide the community with a Python API (for the brave), a GUI to ease the learning curve, and shell scripting capabilities for batch processing, for advanced spectral timing with a correct statistical framework.
Slides from this session
BLOCK 2: Software publishing, impact, & credit
This session focused on using the available infrastructure to better reward software authors and ways to count these valuable research objects. The software contributions that enable much of the results in astronomy are often not recognized, nor considered for reward or promotion. Unlike most of the other sessions in this Symposium, this session had only three short presentations and devoted the rest of the time to an open discussion and Q&A. The session and discussion period was moderated by Rein Warmels (ESO, DE).
Federica Bianco (NYU, US) presented Understanding the Impact of Your Research Software to open the session. She stated that one should always cite software used in research, but “it’s not always obvious how.” She discussed a finding of Howison and Bullard (2015) in their research on software citation in biology articles: open source software is cited more informally than proprietary software. This means that software authors are not accruing credit for their contributions in a way that academia rewards: formal citations. Bianco mentioned Force11, which has published software citation principles, and the now-completed Depsy project, which sought to provide not only citation information on software but also to measure the impact and use of code through other statistics, such as downloads, number of contributors and number of projects reusing the software. Among Bianco’s suggestions for fostering good citation practices for one’s own software were to get a DOI for it and give users instructions on how the software should be cited.
Keith Smith (Science, UK) spoke on Citation of data and software in astronomy: A journal editor’s perspective. Smith said that most scientific advances have their base on previous work, which requires reproducibility. Citations not only enhance reproducibility, they also assign credit. He provided guidelines for citing data and software, sharing bad, better, and good examples, and spoke of a virtuous cycle that will increase reproducibility in addition to the sharing of software and data. Smith mentioned the Center for Open Science Transparency and Openness Promotion guidelines, a policy framework for journals that was developed with journal and community input, and noted that though over 5,000 journals have signed onto them, none of the major astronomy journals have done so. Science’s policies require proper citation of data and software and release of data and software upon publication. Looking forward, Smith sees data and software citation becoming more common, as could be seen with a graph from the ASCL’s dashboard showing the increasing number of citations to its entries, and stated that journals have a role in improving reproducibility and proper citations through policies and editor and referee awareness of changing community standards.
The last presentation of this session was by Alice Allen (ASCL, US), who spoke on Receiving Credit for Research Software. She discussed recent changes in astronomy and in other disciplines that make recognizing the contributions of software authors easier. These changes include new journals, both astro-specific and with a broader focus, specifically for software, policy changes for existing journals, and community resources; these resources include collaborative coding sites such as Bitbucket and GitHub and archival resources such as Figshare and Zenodo. Existing services such as the ASCL have been given new life and are growing; in its Next Generation project, arXiv is improving its support for linking data and code to research. Software citation is captured/tracked/counted by indexers such as ADS, Web of Science, and Google Scholar. Broader efforts to improve reproducibility, citation,
and credit, such as CodeMeta, Force11, WSSSPE, the FAIR principles, and DataCite involve those from many disciplines; the sharing of ideas influences not just those involved in the efforts, but has a greater reach with their aspirational and practical goals and guidelines. Allen shared steps code authors can take to increase the probability of having their software cited correctly and steps researchers can take to improve their articles by including citations for the computational methods that enabled their research, and provided a link to resources mentioned in her talk.
Rein Warmels then opened the floor for discussion. Someone asked whether GitHub would “be there forever?” The point was made that GitHub is not intended to be an archive, and that other services are, so use them to archive your code. On whether to release software, Smith stated that even horrible to read code is better than no code: “it’s hard, painful, and you may hate someone forever… but that’s better than nothing.” Science can provide software information as a supplement, so that is one way to ensure your software is available to support your research findings. In discussing software citations, Allen pointed out that ADS has not been able to automatically track citations to Zenedo DOIs, though that is expected to change soon. The issue of what exactly to cite also came up – should one cite all the dependencies needed to run a particular research code? Neither astronomy nor other disciplines have a way to handle this; at this time, the recommendation is to cite the research software you use well so others will know what your work relied on, and leave it to the software sites to identify the dependencies.
Slides from this session
This is the first in a series of posts on the six-session Software in Astronomy Symposium held on Wednesday and Thursday, April 3-4 at the 2018 EWASS/NAM meeting. Each of the six sessions focused on a different aspect of research software, covering not only specific software packages, but also computational techniques used in data mining and machine learning, open services, software development training and techniques, and getting credit and citations for computational methods. Several sessions included a free-form period in which participants could ask questions, discuss issues, and share information. The last session of the Symposium was a lively moderated discussion among attendees with particular interest in software publishing.
BLOCK 1: Software engineering and sustainability, education for better software, & the ecosystem around Python in astronomy
The first session set the stage for the Symposium, featuring a variety of topics of importance when discussing astronomy research software. Alice Allen (ASCL, US) moderated the session. In the inaugural talk, Software Engineers as Partners in Astronomy Software Development, John Wenskovitch (Virginia Tech, US) opened his presentation with a quote by computer scientist and professor Carole Goble, stating that software is “the most prevalent of all the instruments used in modern science.” This was reiterated by others throughout the symposium. Wenskovitch provided statistics on software use and development activities by academics, among these that 92% of academics use software and 38% spend at least 20% of their time developing software. Research software engineers (RSE) provide guidance to researchers on software engineering and encourage the use of tools that can save academics time and effort in their development efforts. Wenskovitch suggests identifying and using the strengths of each, with the researcher bringing domain knowledge and expertise on the research itself, and the RSE bringing development experience and software engineering expertise. He provided suggestions for ensuring a fruitful partnership; these include using version control, scheduling time for regular and frequent communication, having a prioritized feature list, testing the code thoroughly using unit, regression, and usability tests, and documenting everything.
Mark Wilkinson (DiRAC HPC Facility, UK) spoke next, presenting Research Software Engineering – the DiRAC facility experience. The science requirements for DiRAC demand a 10-40fold increase in computing power to stay competitive, and this increase cannot be delivered solely by hardware. Software vectorization and code efficiency is vital, and RSEs are increasingly important to help with, for example, code profiling, optimization, and porting. DiRAC’s three full-time RSEs are embedded in teams, their time allocated through a peer-review process. Wilkinson showed that the use of RSEs has paid off well for DiRAC, with, for one project, a factor 10 speed-up by optimizing a particular code. The focus on software engineering continued with a talk on Software Engineering Training for Researchers delivered by David Perez-Suarez (UCL, UK). He presented information he had gathered by conducting a quick survey to learn, among other things, what software development training researchers had gotten. His recommendations for training include running or attending training taught by The Carpentries, asking that training be conducted in conjunction with a large conference, such as the American Astronomical Society has been doing for several years, checking to see what software training might be offered by your university, creating your own study group, and contributing to an open source project. James Nightingale (DurhamU, UK) presented a very interesting talk on Test-driven Development in Astronomy. He convinced many in the room that using this technique for developing software will result in better software and less aggravation when coding. He stressed that test-driven development (TDD) is not a testing process, but a development process, and that the code coming out fully tested is a bonus. With TDD, the first task is not to write code, but to write a unit test and then run it to ensure it fails. Only after that do you write the code, and then test it. Through refactoring and testing code, you get instant feedback on whether the code’s functionality has changed, and code design becomes part of the development cycle.
The session then moved on to software sustainability with Bruce Berriman’s (Caltech/IPAC-NExScI, US) talk on Sustaining The Montage Image Mosaic Engine Since 2002. Montage has become increasingly robust and versatile over the years, is embedded in various archives and processing environments, and has been used in other disciplines as well as in astronomy. It has been cited more in information technology literature than in astronomy literature, though uptake of Montage was initially slow. Berriman made the point that design drives sustainability; all Montage releases inherit the design, and each module performs one task. He advocates listening to users and learning from their experiences, and shared his adage that “the grumpier the user, the more valuable the suggestions.”
The last two talks of this first session focused on Python, and covered the growing use of this language in astronomy, the reasons for this growth, the support that is available for the language, and information on one very popular package written in Python. Amruta Jaodand (ASTRON, NL) presented A Walk Through Python Ecosystem, starting with its early development in 1989 by Guido van Rossum at the University of Amsterdam. The advantages of Python include simplicity and natural flow and an extensive, powerful standard library. Strengths of the language include the development of scientific, numerical, and statistical packages and its Python Package Index (PyPi), which enables module and package sharing. Jaodand shared some of the learning materials available for Python, including python4astronomers, and also a lovely Easter egg about Python that is too long to include here and is worth reading. One of the most important astronomy packages is AstroPy, and Jaodand’s talk was followed by The Astropy Project: A community Python library and ecosystem of astronomy packages, presented by Brigitta Sipocz (AstroPy, UK). AstroPy provides software for many common astronomy needs; in addition to the core library, there are many affiliated packages. All of these packages adhere to coding, testing and documentation standards that have been developed by the AstroPy coding community. Sipocz also discussed the community, with members of the team having one or more of its many roles. The number of collaborators continues to grow, and the community welcomes new members, and labels packages that are particularly friendly for a new contributor to work on.
Slides from this session
FRIDAY, 6 JANUARY 2017
Special Session: Perspectives in Research Software: Education, Funding, Reproducibility, Citation, and Impact
10:00 am – 11:30 am
The Moore-Sloan Data Science Environment at NYU and the ASCL have organized a Special Session at January’s AAS. The session, Perspectives in Research Software: Education, Funding, Reproducibility, Citation, and Impact, will be moderated by Bruce Berriman (IPAC, Caltech/Astronomy Computing Today). The session will feature short presentations and will include a discussion period with the floor open for questions and comments, and maybe even a few answers, too. The topics and presenters are :
Tracy Teal (Data Carpentry), Software not as a service
Michael Hucka (Caltech), Finding the right wheel when you don’t want to reinvent it
Lior Shamir (LTU), Reproducibility and reusability of scientific software
Ivelina Momcheva (STScI), Funding research software development
Heather Piwowar (ImpactStory), Capturing the impact of software
David W. Hogg (NYU), The relationships between software publications and software systems
Alice Allen (ASCL), Update on research software citation efforts
That last speaker looks a wee bit dodgy, but the moderator and other panelists are aces! And you, software authors and users, are as always important participants in the discussion. I hope to see you there!
ASCL Editor Alice Allen (that’d be me) was interviewed by Nature Toolbox on the last day of the January AAS meeting. The Q&A appears here.
Earlier this month, Robert Hanisch stepped down as an adviser on the ASCL’s Advisory Committee (AC); we are grateful for his service to the ASCL and thank him for his assistance.
Thomas Robitaille from the Max Planck Institute for Astronomy (MPIA) has joined the AC. He brings with him a wealth of experience as a software author, having developed Hyperion and APLpy; he’s also work on Glue and AstroPy and other astronomy software. We are delighted to have his input!