Georgia Stuart's Introduction

Cyberteams TRECIS
1

I’m Georgia Stuart, a Postdoctoral Fellow at the University of Texas at Austin and a former student at the University of Texas at Dallas. I love HPC, so I’m very excited for TRECIS and Europa! I have been using HPC resources at UT Dallas and TACC for about five years.

I am a mathematician who works on uncertainty quantification for geophysics problems, including seismic inversion, oil spill modeling, and crustal deformation. I’m partial to writing programs in C/C++ (the “expensive part,” at least) and wrapping them in Python for ease of use.

Previously, my workflow was centered around Anaconda for package and dependency management. Recently, however, I’ve been experimenting with containers for use with Singularity on HPC systems. So far it’s going well. I love having all my software and dependencies wrapped into a neat package, and I find Singularity nice to work with. I’m also interested in HPC from a “dev-ops” perspective (thanks Chris Simmons) and how to make reproducible workflows for computational science.

My biggest challenges are

  1. finding compute time (UQ is VERY expensive)
  2. monitoring my experiments as they run since they take so long

Number 1 is the eternal struggle, so I don’t think anything can help with that besides writing more proposals ;), but I would love help with Number 2.

2 Likes
2

Welcome to the community!!!

3

Welcome!

~Joey

4

Hey Georgia,

Thanks for posting!

For 1) we are working on the frontera proposal and I think you said Clint had time on Stampede 2. Do you also have access to LS5? That might help give you some more cycles if you don’t. Either I or Clint could get you LS5 time.

For monitoring your experiments, what type of metrics were you looking for? Are you evaluating some convergence criteria for your sampling? or do you want to visualize posteriors as you sample?

Are we just talking plots or did you have something else in mind?

Longer term, an Open OnDemand App would be very cool but we also want to make sure it’s not the only way to monitor. Not every hpc system will have an OOD portal. However, it might be fun to use containers to spin up your own OOD instance (at TACC for instance) which would then drive your app.

Lots to think about!

Best,
Chris