Building Reproducible Pipelines in R | Will Landau | Data Science Hangout
ADD THE DATA SCIENCE HANGOUT TO YOUR CALENDAR HERE: https://pos.it/dsh - All are welcome! We'd love to see you!
This week's guest was Will Landau, Senior Advisor of Innovative Statistics at Eli Lilly!
Some topics covered in this week's Hangout were building reproducible pipelines with the {targets} package, distributed and parallel computing with {crew} and {mirai}, designing trustworthy AI agents, choosing between Bayesian and frequentist approaches in clinical trials, and maintaining open source software in the age of AI.
One community member asked: When do you reach for a Bayesian approach over a frequentist one in a clinical trial, and how do you explain that choice to stakeholders?
Will's paraphrased answer: The familiar reason is prior information, whether you want to be opinionated about incorporating it through a rigorous prior or techniques like historical borrowing. But he's implemented plenty of Bayesian models on diffuse priors too, because they make early-phase decisions much easier to quantify. Rather than only asking whether a result is statistically significant, Bayesian models let you more conveniently answer the practical question: for example, the posterior probability of observing a large enough effect size, which is incredibly useful when deciding whether to move a therapy to a large Phase 3 registration study. You also don't need frequentist asymptotic approximations like the delta method, you can work directly with MCMC (markov chain monte carlo) and transformations on posterior samples.
Resources mentioned in the video and chat:
🔗 targets Package book→ https://books.ropensci.org/targets/
🔗 targets Package Documentation → https://docs.ropensci.org/targets/
🔗 targets on GitHub → https://github.com/ropensci/targets
🔗 crew Package (distributed computing) → https://github.com/wlandau/crew
🔗 mirai Package (async, parallel, and distributed computing) → https://mirai.r-lib.org/
🔗 Trusted Mini-Agents Guide (Will Landau) → https://wlandau.github.io/trusted-mini-agents/
🔗 RTMB Package → https://cran.r-project.org/web/packages/RTMB/index.html
🔗 Will Landau's targets Workshop at R/Pharma 2023 → https://www.youtube.com/watch?v=cyF2dzloVLo&list=PLMtxz1fUYA5C0csy8-wTfFmRYQEB30vr3&index=12
🔗 Will Landau & Charlie Gao joint talk at posit::conf(2025) → https://www.youtube.com/watch?v=s_cBi155ZIQ
🔗 Data Science Lab episode with Charlie Gao on mirai → https://www.youtube.com/watch?v=Vf_jmY1tMWo&list=PL9HYL-VRX0oSeWeMEGQt0id7adYQXebhT&index=1
🔗 "AI and the Invisible Newcomer in Open Source" by Mara Averick → https://blog.stdlib.io/ai-and-the-invisible-newcomer-in-open-source/
🔗 Kahn's Algorithm for Topological Sorting → https://medium.com/@robhernandez5/kahns-algorithm-for-topological-sorting-26f29f2eaf48
🔗 XKCD: Dependency comic → https://xkcd.com/2347/
🔗 MLU-Explain (visual machine learning explanations) → https://mlu-explain.github.io/
🔗 Posit Academy → https://academy.posit.co/
â–º Subscribe to Our Channel Here: https://bit.ly/2TzgcOu
Follow Us Here:
Website: https://www.posit.co
Hangout: https://pos.it/dsh
The Lab: https://pos.it/dslab
LinkedIn: https://www.linkedin.com/company/posit-software
Bluesky: https://bsky.app/profile/posit.co
Thanks for hanging out with us! 💛
Timestamps:
00:00 Introduction
04:58 "Can you talk about the targets package?"
13:21 "You're doing great work to try to separate out that development support stuff from your day-to-day job and personal time and life. How the heck do you do it?"
16:57 "As someone who's primarily worked with tools like SnakeMake and Nextflow, why should I use targets, or what are some benefits of using targets over something like Nextflow?"
20:37 "How has Gen AI affected you as a package author and maintainer, mostly from other people submitting issues or PRs, and less about how you use it personally in your own work?"
25:06 "Do you want to talk a little bit about integrating Mirai to do things even faster?"
29:10 "How do you communicate uncertainty to people who want a single number?"
31:24 "What does an average day or week look like for you? How is your time divided between development, consulting, etcetera?"
34:00 "Have you had feedback from people using targets for agentic LLM workflows, for testing parameterization changes of RAG ingestion and associated changes to how the LLM instructions are dealing with that, or just anything in that creating and testing agentic workflows space?"
37:15 "When do you reach for a Bayesian approach over a frequentist one in a clinical trial, and how do you explain that choice to stakeholders?"
41:32 "Which Bayesian MCMC tool do you use? Do you use Stan, or the macros to get to Stan?"
45:56 "Is there a piece of career advice that you wish you knew before you started in data science, or anything that you think would be helpful for everybody else here?"
This week's guest was Will Landau, Senior Advisor of Innovative Statistics at Eli Lilly!
Some topics covered in this week's Hangout were building reproducible pipelines with the {targets} package, distributed and parallel computing with {crew} and {mirai}, designing trustworthy AI agents, choosing between Bayesian and frequentist approaches in clinical trials, and maintaining open source software in the age of AI.
One community member asked: When do you reach for a Bayesian approach over a frequentist one in a clinical trial, and how do you explain that choice to stakeholders?
Will's paraphrased answer: The familiar reason is prior information, whether you want to be opinionated about incorporating it through a rigorous prior or techniques like historical borrowing. But he's implemented plenty of Bayesian models on diffuse priors too, because they make early-phase decisions much easier to quantify. Rather than only asking whether a result is statistically significant, Bayesian models let you more conveniently answer the practical question: for example, the posterior probability of observing a large enough effect size, which is incredibly useful when deciding whether to move a therapy to a large Phase 3 registration study. You also don't need frequentist asymptotic approximations like the delta method, you can work directly with MCMC (markov chain monte carlo) and transformations on posterior samples.
Resources mentioned in the video and chat:
🔗 targets Package book→ https://books.ropensci.org/targets/
🔗 targets Package Documentation → https://docs.ropensci.org/targets/
🔗 targets on GitHub → https://github.com/ropensci/targets
🔗 crew Package (distributed computing) → https://github.com/wlandau/crew
🔗 mirai Package (async, parallel, and distributed computing) → https://mirai.r-lib.org/
🔗 Trusted Mini-Agents Guide (Will Landau) → https://wlandau.github.io/trusted-mini-agents/
🔗 RTMB Package → https://cran.r-project.org/web/packages/RTMB/index.html
🔗 Will Landau's targets Workshop at R/Pharma 2023 → https://www.youtube.com/watch?v=cyF2dzloVLo&list=PLMtxz1fUYA5C0csy8-wTfFmRYQEB30vr3&index=12
🔗 Will Landau & Charlie Gao joint talk at posit::conf(2025) → https://www.youtube.com/watch?v=s_cBi155ZIQ
🔗 Data Science Lab episode with Charlie Gao on mirai → https://www.youtube.com/watch?v=Vf_jmY1tMWo&list=PL9HYL-VRX0oSeWeMEGQt0id7adYQXebhT&index=1
🔗 "AI and the Invisible Newcomer in Open Source" by Mara Averick → https://blog.stdlib.io/ai-and-the-invisible-newcomer-in-open-source/
🔗 Kahn's Algorithm for Topological Sorting → https://medium.com/@robhernandez5/kahns-algorithm-for-topological-sorting-26f29f2eaf48
🔗 XKCD: Dependency comic → https://xkcd.com/2347/
🔗 MLU-Explain (visual machine learning explanations) → https://mlu-explain.github.io/
🔗 Posit Academy → https://academy.posit.co/
â–º Subscribe to Our Channel Here: https://bit.ly/2TzgcOu
Follow Us Here:
Website: https://www.posit.co
Hangout: https://pos.it/dsh
The Lab: https://pos.it/dslab
LinkedIn: https://www.linkedin.com/company/posit-software
Bluesky: https://bsky.app/profile/posit.co
Thanks for hanging out with us! 💛
Timestamps:
00:00 Introduction
04:58 "Can you talk about the targets package?"
13:21 "You're doing great work to try to separate out that development support stuff from your day-to-day job and personal time and life. How the heck do you do it?"
16:57 "As someone who's primarily worked with tools like SnakeMake and Nextflow, why should I use targets, or what are some benefits of using targets over something like Nextflow?"
20:37 "How has Gen AI affected you as a package author and maintainer, mostly from other people submitting issues or PRs, and less about how you use it personally in your own work?"
25:06 "Do you want to talk a little bit about integrating Mirai to do things even faster?"
29:10 "How do you communicate uncertainty to people who want a single number?"
31:24 "What does an average day or week look like for you? How is your time divided between development, consulting, etcetera?"
34:00 "Have you had feedback from people using targets for agentic LLM workflows, for testing parameterization changes of RAG ingestion and associated changes to how the LLM instructions are dealing with that, or just anything in that creating and testing agentic workflows space?"
37:15 "When do you reach for a Bayesian approach over a frequentist one in a clinical trial, and how do you explain that choice to stakeholders?"
41:32 "Which Bayesian MCMC tool do you use? Do you use Stan, or the macros to get to Stan?"
45:56 "Is there a piece of career advice that you wish you knew before you started in data science, or anything that you think would be helpful for everybody else here?"
