R in HPC Environment

Instructor: Mark Lowerison (Community Health Sciences)

Using R and a variety of other technologies, I plan to demonstrate a multi-agent parameter search simulation project. We will be looking at differences in performance of two randomization strategies confounding mitigation in clinical trial settings.

The technical approach:

1. Generate a simulation space/parameter set to be analyzed.
2. Write a simulation agent to consume parameter sets.
   1. This agent will fetch a parameter set.
   2. Generate a synthetic data set.
   3. Model this dataset and assess confounding between a key covariate and a treatment effect.
   4. Extract results from the R model output and push them back to a results store.
3. Write a minimal dashboard to monitor the simulation under way and summarize some results as they progress.

Target audience: Attendees will walk away with working code samples and understanding of how to leverage distributed computing infrastructure to support containerized development of multi-agent simulation projects in R (and friends).

Duration: 3 hours

Level: intermediate to expert

Prerequisites: In addition to R, this presentation will use a little bit of Python, Git, Docker, Singularity, bash and Slurm. This course assumes a basic familiarity with at least some of these tools. In particular, some previous experience with R (or DataFrames in pandas) would be extremely valuable.

Laptop software: All attendees will need to bring their laptops with wireless access and with a remote SSH client installed (on Windows laptops we recommend the free edition; on Mac and Linux laptops no need to install anything for ssh). It will also be important for attendees to have Git and Docker installed on their laptops.