Statistical Software Engineering

We have all these SAS macros, what do we do with them? Yeah, no, I think that’s, that’s really key to really scale up the, the number of users and the number of developers, you know, I [00:17:00] mean, even, even in a bigger company like Roche, I mean, it wouldn’t make, it just wouldn’t make sense anymore nowadays to say, well, for our 1000 plus, you know, statisticians working in technical development, we just develop everything in house.
It just wouldn’t make sense because you don’t have enough users and you’re, especially, you don’t have enough developers, right? Right. Okay, you can always, let’s say, buy developers from vendors and something, but it just, it just gets done too expensive, right? It’s much better, I think, to see that, that, that so many things can be jointly developed between companies sometimes also academic parties are interested to use certain tools then and so on.

You just scale up your user base, you scale up your developer base, and everybody kind of saves cost. All right. And everybody kind of benefits from the better quality that you get. And again, as mentioned before, and it makes so much sense to base many of these, for example, workflows on really kind of public tools that are used, that are used for [00:18:00] completely different things, right?

Outside of statistics, outside of pharma. Right. I mean, these GitLab or GitHub workflows, they’re used for completely different stuff. Right. And then, and then you have this benefit. You have much more, much bigger user base. You have much more professional and bigger developer base. Yeah. Yeah, definitely. You talked about a couple of different tools and what are kind of essential tools that people should use to, or that you recommend using for that help with all the software engineering part?

Yeah, that’s a great question. So, so what are the tools? Yeah. I mean, so, I mean, what is, what is very critical? And I think Hinted to this before already is that people learn to use Git. Okay. Git is the, is the dominant version control tool nowadays. It [00:19:00] hasn’t always been like that. Of course, before there were other systems and maybe in the future there will be similar other systems again, but at the moment it’s Git.

And I think that’s what we also see internally as a, as a critical bottleneck when we do trainings. for onboarding new colleagues or colleagues onto this new kind of systems where we use R and, and, and other open source things. It’s, it’s learning Git, right? So definitely invest time visit some workshops visit the websites, learn Git.

And, and honestly, I mean, if you manage to get to this kind of programming or statistics job, then you will, you will also manage to learn Git. It’s not, it’s not harder. Right. It’s still much easier than all of the math and stats and whatever you learned before. Right. But it just takes, it just takes a bit of, you know, it’s like, I don’t know, learning something that is not so much fun, but then you need to have it.

Right. It’s one of those things. Right. And actually when you, [00:20:00] when you can do it, actually, it’s all, it can also be a little bit fun. Okay. Git is like critical. What else is critical, I would say is to get familiar with, with you know, good coding habits, right? Starts from simple things like so called clean code rules, right?

Which is about, you know, how can we, how can we name things that people understand what they mean? You know, don’t name your variables X and Y and Z, name something that they actually mean, right? For example, cost and what a value and probability or whatever, right? That is and it goes to things like. You know, making, making sure from the indentation and, and styling of your code, people can actually, it’s kind of consistent and there’s many good tools, right?

And R, there’s Lint, R and Styler to basically help you style and check your style. And so there’s many, many tools available and of course, not just for R, [00:21:00] so there’s, there’s many of this kind of habits, right? And so on. And, and the other thing, which, which I think is really helpful is like getting into this habit of code review, right?

So that you don’t just sit alone at your program and you just do everything alone. And then. You have a lot of risk that you make some mistakes. You have a lot of risk that a second person will not understand this. You have a lot of risk that you yourself will not understand in two to three years from now, that you say, well, it was basically, it should be like a teamwork, right?

It should be like a teamwork project. Of course, not for some super experimental stuff. You just quickly play around, but for, let’s say for production work, anything that’s kind of goes into any kind of decision making, you know, and that can be the sample size for the next study, that can be the submitted table for a regulator, that can be anything like that That is, that is [00:22:00] a little bit important, right?

At least just get into the habit of having somebody else look at your code. And, and, you know, of course, then also take some practice to give comments as a reviewer, right? You need to be nice ask questions. Maybe there was a good reason that somebody coded like this. But there will be a lot of good ideas from you as a reviewer how to improve the code, right?

And then you make, and with that you really make sure there was at least two pairs of eyes on this code, and you have a much better chance of avoiding mistakes of, of having this readable also in two, three years by somebody else, right? So, so I would say, yeah, version control, Clean code rules and code review is some, some key to start components.

Okay. Yeah. And it’s great. That says actually some tools that help you with. Becoming better in these habits, that’s, that’s pretty cool. Now we talked a lot about R here. And of course there’s another software that many of us are [00:23:00] using and that is SAS. So, in how far can we apply all these learnings that we have, the lessons that we have for R also to SAS?

Thanks. Yeah, great question. I would say in principle, you know, most of these habits and skills and tools apply to any programming language. And if we, and if we talk about SAS as in terms of the SAS Programming language, then it’s definitely also applies, right? And it equally also applies to others like Python, Julia, C and whatever, right?

I think I’m personally not so Experienced in using SAS, at least for the last 15 or years, I did, I did, I did, I mean, I did a couple of things in an internship, like 16 or 17 years ago. And I mean, to me, a little bit, the question would be. Is, is, is it possible to use the SAS language and then [00:24:00] still apply, for example, things like automated workflows and so on, right?

What we talked about before that I would not be so sure, but in print, but in principle, and that’s also what we, that’s also what we, let’s say, support here, for example, in Roche with our new data science platform called ocean that, that a lot of people have been building up. And I think that looks very nice is that we actually, you know, it’s a multilingual platform.

It’s not just art. There’s a lot of SAS still. Right. There’s also the possibility to use Python and Julia and so on, and maybe other things in the future. And and of course, you know, many, many, many studies, especially for molecules that have started development years ago, they started in SAS, and it would have been too much work to not translate all the code to R.

So they continue in SAS and they will still continue to live their life cycle with SAS. And the SAS is definitely supported by this new platform, Ocean, and You know, there’s certain ways to automate things [00:25:00] and maybe it’s using a little bit different, let’s say scripting language or whatever, but it is also ways to do that, right?

And there’s also ways to do that well. And, and maybe, and maybe I also want to say, you know, I think, I think sometimes that also statisticians can learn from statistical programmers in the sense because statistical programmers have been using things like double programming. And maybe code review as well for, for a long time, actually.

Right. And they have been talking about things like, okay, let’s make certain things standard. It’s not everybody, right. Things differently. Right. So, so I think it’s, we also should appreciate this experience, especially when you work in a pharma company and, you know, talk with your statistical programmer colleagues and see what they can, what you can learn from them.

Right. Yeah. Yeah. Now, how can people learn more about all these kinds of different things? How can people learn to become better in terms of software [00:26:00] engineering? Yeah. Great question. So the one thing is, I mean, the one thing I will, I will make a plug is OpenStatsWear. org. So that’s a scientific working group mostly comprised of statisticians, but also some programmers.

So different, different backgrounds, but everybody’s interested in kind of driving forward and, and contributing to the kind of open source package environment in, in statistics in particular. So what we do is we build packages on the one hand. We put them together in work streams to different topics, different packages.

And we also try to spread the word about best practices. So we did a couple of, we actually ran a couple of workshops already. So we did a little bit of a world tour last year. The last, the previous one, the last one I ran was in Zurich. And we will hopefully have more in the future. So keep definitely watching for those things to happen.

And, and if you already. Interested in package development, please reach out to us at openstatsweat. org if you want to get [00:27:00] involved. That’s one, let’s say, group of people that are really interested in this topic and also want to improve the topic. We will also try to, to make maybe some online versions of this kind of workshops now in the future to, to scale this a little bit beyond the face to face experience, scale this a little bit globally as well.

So, so keep looking for that. And and maybe, maybe a more, how to say, maybe a more general advice. you know, have a look, I would say, I would recommend to have a look where there are opportunities in your organization to do this kind of thing. And look for those people that are already doing this kind of thing in your organization, right?

Because whether it’s in academia, there’s many research software engineering or RSE groups. popping up in universities and academic research centers. So maybe have a chat with them, maybe see how you can get involved in a project together and that you can actually learn from them. And similar in, [00:28:00] in companies, right?

Pharma companies. But what we have seen through OpenStats for that August that actually many pharma companies also now have special teams that are working on packages and software. And that can be more in the traditional stats application that can be more in the machine learning applications.

But if you’re interested to learn more then definitely try to get in touch with them. Maybe they do some presentations, maybe they even have the opportunity for rotations, right? Temporary kind of allocations into their groups. And, and we have seen that at least internally here in Roche with the statistic engineering team that I build up, this has been working really nicely, where we had, we had people from all over the world, from the Roche organization, Joining part time into the team and you just learn so much when you can do the work hands on and Get it reviewed and talked about with people that are experts in this area.

Right. And, and that, that will get you the real deal [00:29:00] and it will be a lot of fun. Awesome. Thanks so much. So OpenSTED software is of course, great, great thing. Now there is case studies where. Yes, it’s open software, and then you want to kind of implement it within your company and get it kind of validated and all these kind of different things.

And RPECT is such a tool that has this kind of approach. Can you tell a little bit more about RPECT and also how that relates to your company that you’re now starting? Yeah, thanks for the question. That’s a great question. So, so RPACT is our package specialized for adaptive clinical trial design and analysis and simulation, and it also includes things like group sequential designs, which, which many statisticians will be very familiar with.

And I think has been a great example of. a business model, how we can, [00:30:00] you know, make this, make this work that on the one hand, we have a super high quality implementation as an open source package. Right. So our practice freely available from, from ground and get up where, and, but on the other hand, how can we then make sure the people that are developing this package actually also.

get the money to do it, right? And it has been kind of a crowdfunding approach where, where companies and organizations kind of chip in to jointly kind of finance the development of this package, right? I think that’s a great story that I’ve, you know, seen a little bit from the beginning when I was on the rush side, actually years ago.

The first interactions and then now I’ve, I’ve also seen kind of maturing over time and where I think you know, there can be hopefully more examples of this and maybe also other business models to, to make this work. And of course, I’m also super excited that you know, a while ago When, when I thought about, okay, what to do next, and especially [00:31:00] with the relocation to, to Asia and Taiwan, what, what am I going to, going to work there?

And I’m so super excited that we chatted with you know, Gernot from, from ARPACT, and then we kind of line up this kind of joint venture with them. To, to help them, you know, build out further the kind of great work that I’ve been doing around RPACT and, and maybe also other products, right. And this is basically the new Arconis.

So spelled R C O N I S. You can see more at arconis. com. That’s a new kind of company that, that we have been starting now. And yeah, it’s super exciting. Yeah, definitely check out this homepage. There’s some really exciting stuff happening in the future. And yeah, stay tuned, follow Daniel Friedrich, Gernot and last but not least Carrie Daniel’s wife said is.

Who’s also a statistician and who is joining there as well. [00:32:00] So thanks so much for that awesome discussion about what we can learn from Google, about software engineering, about tools, ways we can do better and actually get more out of all the things that we’re doing in a Yeah. Yeah. And basically really automate some mundane stuff so that we can concentrate on the more important things.

What’s your final takeaways that you want the listener to have from this podcast episode? Yeah, first of all, thanks Alexander for the invitation. Just having listened to a couple of episodes of the effective statistician podcast before, I feel very happy and humbled to be here. So, so thanks for the opportunity.

And I think what, what to keep in mind is I would, I would hope that many statisticians start to take software engineering and coding a little bit more seriously. So, you know, try [00:33:00] maybe next time you code something, try to think about think about polishing it a bit more, you know, think about creating a piece of art.

It’s not just to get something done quickly. It’s, it should be something nice. And, you know, it should be something like, I don’t know, you’re writing like a love letter to you or to a partner or something. It should be something looking nice and something nice to read through. And if we all, I think if, if people invest a bit more, into the quality of, of the code they write.

I think we, it can help everybody to be more efficient, avoid problems, avoid bugs and it will be more fun. And so, yeah, so maybe, maybe try out next time. Awesome. Thanks so much for it. Thanks, Alexander.