Labou "Data Science and the Library at UC San Diego"

Editor's Notes

• #2: Hello, my name is Stephanie Labou and today I’m going to be talking with you about data science and the library at UC San Diego.
• #3: I want to start with some background information, since it will help contextualize my perspective. I’m at UC San Diego, which is a large R1 university – meaning we are doctoral granting with very high research activity. We have a medical school and a business school and about 39,000 students. UCSD had a Data Librarian for decades, but the position was recently rebranded as Data Science Librarian. I’ve been here almost two years and I’m the inaugural “data science librarian”. I do want to mention: I don’t have a library degree and this job is my first job working in a library. But! I do have a master’s degree and before this job, I worked for 3 years as a data manager and research assistant with a large interdisciplinary environmental research group. So I have lots of experience working with data and – crucially – with scientific programming.
• #4: So this is today’s topic. I wanted to highlight a few phrases in particular because these guided how I put together this talk. Of all the topics, I want to focus on two in particular: First, how can information professionals be well-versed in data science skills in order to fill a support role, and second, what tools do information professionals need to work with data moving forward? I’ve split these into the two components – outward vs inward – because I think the skill sets are complementary, but not necessarily the same.
• #5: To start, I want to take a step back because all this is predicated on the term “data science”. So what is “data science”? Well, it’s got a lot going on. You’ve maybe seen one of these types of data science venn diagrams and you can see that data science is a large, often poorly defined, multidisciplinary, and ever-evolving field. https://www.kdnuggets.com/2016/10/battle-data-science-venn-diagrams.html
• #6: To be blunt, there’s a lot of hype about data science. I see two flavors of data science: the “hard core” stuff like AI and deep learning and the aspects of data science that are really like computer science engineering. And then there’s the rest of data science – which, to be clear, I love! – which are what I think more of as modern data literacy skills. This is being able to work with data. You may have heard the statistic that 80% of a data scientist’s work is data cleaning, and it’s true. I don’t mind the term “data science” for these kinds of skills. I think they’re incredibly important and they are, in a real sense, the “science of data”. This is what I’m going to focus on in this talk. I wanted to touch on this briefly because I also think the term “data science” has a lot of baggage. It can intimidate people. It can be stressful to think “oh no, now I’ve got to learn machine learning and AI and it seems like such a big leap”. And I think being clear about what we mean when we say “how are libraries using data science, what should librarians know about data science” is not just helpful, but necessary in terms of setting expectations and goals.
• #7: So let’s start by talking about data science as a support area – how can we support students and researchers.
• #8: For this, I’ll talk about my role specifically. What kinds of questions does the data science librarian get? Well, a lot of them are about finding data. Because what does data science methodologies need? Data! For the other type of question, it’s really about how do we extend the support we provide for data, to data science.
• #9: And the common thread here is computing. These questions – coming from pretty much all disciplines on campus – are about using data in compute-heavy ways. They’re about working with data in R or Python, or working with non-traditional data formats like JSON or HTML of netCDF, which was mentioned in the webinar last week – formats that we new to people or disciplines that are used to working with spreadsheets. It’s questions about API use and web scraping – about accessing and leveraging the massive amounts of data that are available out there. It also means that sometimes, a traditional laptop isn’t going to cut it. The library at UCSD has a Data & GIS Lab and I see it as filling the space between “I can do what I need on my laptop” and “I need a supercomputer”. This is for large – even TB – datasets and our computers have more memory and processing power than a laptop. So sometimes the “support” is providing hardware, as well as software and software help. It is this niche that’s not computer science, but is about reproducible research, research automation, scientific programming. Which means I also spend a lot of time talking about best practices and how to get started with these things, because usually people aren’t coming at this with a background in computer science. Often, they’ve never taken a single programming class, or maybe they’ve taken a quick bootcamp.
• #10: Ok! So, considering that that’s my purview, what skills do I need to do this? Obviously, I needed to have a strong grasp of the entire data life cycle. It has also been crucial for me that I had deep knowledge of at least one programming language. I consider myself quite experienced with the software program R, and I know enough to be dangerous in Python and Stata. But, a lot of that additional platform knowledge entails a lot of Googling – I know what I want to do, and I understand the conceptual framework or order of operations to get there, in a programming sense, but I may not know the exact syntax. But I can Google that. And I’m self-taught when it comes to programming. I took a stats class in grad school that used R, but everything else I learned on-the-job in my previous position. There’s nothing like getting thrown in the deep end to make you learn fast. Plus, learning project-by-project also meant that I picked up the skills that were most useful first, rather than starting from fundamentals. Whether this was a good thing is debateable – I know I’m missing some building blocks of basic computer science knowledge – but it rarely causes problems.
• #11: So, do we all need to learn data science to be able to provide data science support, in this sense? Well, probably not, but it depends on what level of support you want to be able to provide, which in turns depends on what your patron needs are. We’re a large, STEM-heavy campus, with a data science institute and major, so it makes sense for us to be able to provide this deep level of support, which includes code support from someone who is not only conversant, but experienced with at least one programming language. But, being conversant is often enough. The important thing is the conceptual framework of how to approach a problem. For instance, if I need to present data, how can I get data from format and structure of type X to type Y? Breaking this down into steps: first, I need to convert my character dates to date formats, I want to have columns of ABC which are currently in row format, etc. Thinking through the workflow is what helps the most. And this is something that comes from experience working in a programming language but a little goes a long way. Long vs wide data, the concept of grouping data, etc. – a lot of basic data literacy, but in this programming/data science context. And for other librarians, I would say that this is just one more reference area. As a reference area, it’s more about knowing where/what to search: knowing the names of some common platforms, maybe the names of some packages, and where to find help.
• #12: So, this is the big question and again, depends: we don’t all need to learn everything, but it is helpful to know a bit. And the level of support your organization can provide will increase with more in-house knowledge. The first option is, of course, skilling up existing employees. I just finished saying that we don’t all need to become full-fledged data scientists, and I believe it! But, I do think having the exposure to it can definitely help provide more in-depth support for patrons. This is a popular topic, not just within data science, but across all domains. So how can libraries scale up their support of data, to support for data science? The Carpentries organization – an international organization with free curriculum online – has a curriculum for Library Carpentry, which is these kinds of skills for librarians. It covers a variety of different platforms for automating tasks, including R and Python, as well as command line and others, as well as how to get into that mindset of working with data at scale in an automated fashion. We’ve run one of these workshops here recently and it was quite popular. Another option is my experience: bringing a non-library trained professional in. It has been an amazing fit for me – libraries is clearly where I belong – but I know this isn’t always a popular option. But until MLIS curriculum changes to incorporate more of this – and if it should is another conversation; no one program can prepare one person with every skill – this is a definite option to hit the ground running with providing a deeper level of support. I had that in-the-field skillset so I can work with students and faculty at a higher level, from recommending specific packages, to teaching workshops, to reviewing code. The third option I see is outsourcing these duties to another group, likely one on campus. This may be the IT group, or a research facilitator group, or another group. This does assume though, that these groups (a) exist and (b) are willing to take this one, which may be not the case. I’m fortunate that at UCSD we have a vibrant campus-wide collaboration for all things data science and we can each tackle a component that fits most naturally. For instance, central IT runs an online virtual machine, in essence, that students can use to run GPU-intensive calculations. And honestly, I’m relieved that that’s not something I need to worry about – it exists and it’s not my department. But it does exist!
• #13: Ok, so moving on to data science within the library. This is less about supporting patrons and more about using data science to enhance our own work.
• #14: There’s been a lot of talk about data science methodologies in libraries, but – and correct me if I’m wrong – I haven’t yet seen more than case studies from more university libraries. That is, no one has yet gone 100% data science and revamped their entire workflows. But, please let me know if I’m wrong here! What I have seen other groups do, and what our library has done, is implement “data science” methodologies for certain projects. So for example we’ve created Python scrips to automate parts of metadata ingest into our internal system for certain projects. The cataloguers seem quite keen on using OpenRefine, which is an open source platform for automated data cleaning. We’ve talked about how we would like to figure out how to automate scraping DataCite and maybe leverage API capabilities more than we currently do. We’re definitely still in the early stages – we have some folks in-house who know Python and can do these types of things, and I’ve also had some of the students in the Data & GIS labs work on some Python scripts for us as well. I’d love to get some data science majors hired in at some point, but that’s for the future and for specific projects.
• #15: The main take away, I think, from what we’ve done so far, is that using data science in libraries should be a case of using new methodologies because they solve a problem faster or better. Not because they’re flashy and everyone is talking about it and the higher ups are suddenly expecting us to “do data science”. It’s really been about automating manual processes, or talking about how we could better assess our own metrics, or better display information. It is targeted and it is on a case-by-case basis by people who feel comfortable with it and are eager to learn and implement these projects.
• #16: And from other libraries and universities, I’ve seen some great “collections as data” projects, which entail making existing collections more accessible for data science topics. So using a sprinkle of data science and FAIR principles – findable, accessible, interoperable, and reusable – to make existing collections more likely to be used for these data science types of projects. Especially for natural language processing or text mining, this is a huge opportunity. I also think we should be thinking a lot about data as collections. I said it before and I’ll say it again – what does data science methodologies need, in every domain? They need data! We have a research data collection here at UCSD, as well as data we have purchased or licensed, and I want to see those collections, or guide, as high value and high use, even more so than they currently are. This is a chance for reuse of data to really speed up in certain fields and I want the library to be at the center of it.
• #17: I want to close with a few takeaways. Adaptability and flexibility will be key, in terms of data science and libraries moving forward. There’s a lot of hype and focus on specific platforms, or languages, or packages, or what have you, but these things will absolutely change. What won’t change is best practices. How to manage data and code and information. How to structure data and projects. Where to find data, how to cite it, when to reuse or not reuse data. These are all areas that absolutely fall within the library’s purview and where the library can excel. This is modern data literacy and by focusing on those aspects, the library can not only remain relevant, but provide much-needed guidance. This also fits when talking about infrastructure: hardware, software, and people. We’ll have to adapt and we’ll have to grow. It can be scary. But it can also be a good opportunity. So planning forward when talking about what spaces we want, what staff we want, and what hardware/software capacity and capability we want. Finally, collaboration. I’m the point of contact for data and data science, but I work closely with our subject libraries, other groups in the library, and our library IT. I work with campus IT, our research facilitators, and the Halıcıoğlu Data Science Institute here. Data science is way too big – not just in terms of TB scale and hardware, but also that it’s now in basically every domain across campus – for any one person or group to handle. Remember the Venn diagram? It’s got a lot going on! Collaboration is key and I think a lot of the reasons my position in particular has been so successful is that there was a network of people I could work with, which has been invaluable.
• #18: Thank you very much for listening. I’d be happy to take questions now and you can of course email me with questions about this or anything else related to data science. My email is here: slabou@ucsd.edu. Thanks again!