Combinatorial Databases
Contents
Combinatorial Databases¶
In late 2021 Robert Craigen started a discussion with a number of researchers in “Hadamardish orthogonal stuff in combinatorics” about the need for a “more systematic, comprehensive and permanent reference system for matrices, sequences, and all sorts of miscellany pertaining to Hadamard matrices and the various generalizations, variations and spin-offs”. The discussions identified a large number of existing online repositories with wildly different implementations and user interfaces, but no one path forward.
This book is the result of an extended investigation of options. It may or may not turn out to be a good way to proceed, but gives one possible step towards Craigen’s goals, and will hopefully help push the discussion forward.
Mathematical Repositories, a very brief history¶
The oldest known mathematical data repository may be a 4500 year-old tablet from the Sumerian city of Shuruppag. See [CKCF+03] for a history of mathematical table making over the millenia, through logarithm, actuarial and astronomical tables up to spreadsheets.
For most of the 20th century, mathematical data was published either in book form (for example [Col10]) or as microfiche supplements to journal articles. Later, online repositories began to appear, but were often scattered in different places, often made by people without knowledge of other overlapping material. URLs often changed, leaving broken links behind. Finding and categorizing everything available on a given subject can be challenging; see for example [SBFM20] for a thorough examination of graph theory sites, and [BervcivcV20] and [FGorgen22] and their associated repositories DiscreteZoo and MathRepo for a look at some different types of online mathematical databases aspiring to FAIRness.
Many other STEM areas, such as biology, geophysics and crystallography, routinely deal with large datasets, and so are far ahead of mathematics in adapting to new options for presenting their data. In 2014 the Joint Declaration of Data Citation Principles laid out principles for data citation, and papers such as [SCC+15] and [FCG+19] gave roadmaps for achieving these principles.
There are a number of large initiatives aimed at advancing the handling of data in mathematics, largely in Europe. OpenDreamKit (Open Digital Research Environments Toolkit for the Advancement of Mathematics) ran from 2015 to 2019. MaRDI, the Mathematical Research Data Initiative, is a German initiative that started in 2021.
FAIR Data¶
In recent years the importance of making research data in all areas of science and technology available has been recognized as crucial, resulting in the FAIR data principles: data should be:
Findable
Accessible
Interoperable
Reusable
These guiding principles were detailed in [WD16]:
To be Findable:
F1. (meta)data are assigned a globally unique and persistent identifier
F2. data are described with rich metadata (defined by R1 below)
F3. metadata clearly and explicitly include the identifier of the data it describes
F4. (meta)data are registered or indexed in a searchable resource
To be Accessible
A1. (meta)data are retrievable by their identifier using a standardized communications protocol
A1.1. the protocol is open, free, and universally implementable
A1.2. the protocol allows for an authentication and authorization procedure, where necessary
A2. metadata are accessible, even when the data are no longer available
To be Interoperable:
I1. (meta)data use a formal, accessible, shared, and broadly applicable language for knowledge representation.
I2. (meta)data use vocabularies that follow FAIR principles
I3. (meta)data include qualified references to other (meta)data
To be Reusable:
R1. meta(data) are richly described with a plurality of accurate and relevant attributes
R1.1. (meta)data are released with a clear and accessible data usage license
R1.2. (meta)data are associated with detailed provenance
R1.3. (meta)data meet domain-relevant community standards
This paper [BKR20] looks at how FAIR data principles apply to mathematical data, and propose a further principle they call “deep FAIR” data, and give a sampling of datasets here.
In the next sections we discuss what these principles mean in the context of combinatorial data repositories, and how they influence the design choices of this book.
Findable¶
There are many online repositories of mathematical data, but locating them can be difficult, and often links are broken when a website has moved or just disappeared. So establishing a permanent location is critical.
GitHub is one place that code and data can be deposited, but it is not designed for permanent preservation. A number of repositories that are designed to be permanent and discoverable are available, such as Dryad, Figshare, and Zenodo. They all assign a record a DOI (Digital Object Identifier), by which it can be uniquely identified and located.
They have various pros and cons, and new ones are regularly appearing. I decided to go with a GitHub repo archived at Zenodo.
Note that the author’s website [Gor22] still exists, and provides a friendlier interface for users who only need a few of its combinatorial objects. For now this book is a supplement, but the long-term plan is for it, and others for difference sets, covering designs and Steiner systems, to replace it entirely.
Adding metadata so that tools like Google Dataset Search can find the repository is important. There are also directories of research data repositories, such as https://www.re3data.org and https://fairsharing.org. Making this data findable is still a work in progress.
Accessible¶
How to present data is another question. Jupyter Notebooks are a way to present data and code together, so that a user can quickly manipulate it. Jupyter books are a way to add material beyond the notebook. In [SF21] a Jupyter Book is used to prepare data, a paper, and slides for a talk all in one package.
One way to interact with a Jupyter book is to install the Jupyter Book software, the source for the book itself, and then build it. See [Com20] for information about how to do that.
That’s more work than most casual users will be willing to do. An easier alternative is to use BinderHub, which can be set up to execute a Jupyter Book on GitHub. See the Launch into interactive computing interfaces section of [Com20]. Click on the launch button above to try it out.
Interoperable¶
Presenting the data as a CSV file seemed like the most direct approach. Simple Python code is included to do basic processing, so that the data can be examined without additional work, and can be used with other people’s code with a minimum of effort.
The goal was to keep the data and code as simple as possible. This dataset is fairly small, which is why it was chosen to start with. The covering designs database is several gigabytes. That amount of space is not a problem, but putting it all in a Jupyter Notebook and getting good performance may be more challenging.
Reusable¶
When you store your data on a repository, you will need to specify a license for use. Creative Commons has a number of options:
CC BY: reuse is allowed, but must cite the creators
CC NC: no commercial use
CC0: nearly public domain
This paper In [LM20] looks at questions about
sharing research data.
These workshop slides [KS17] give a good overview of the
steps and options involved in publishing research data.
This book and its data are released under the CC BY license.