16 Bits from the Winter Hackathon
1. New year, new hackathon! The January 30–31 event was Computing’s 23rd hackathon and the 1st scheduled in the winter season. The directorate will hold four hackathons in 2020.
2. Hackathon responsibility rotates among different Computing divisions. With a new hackathon on the calendar, the Data Science Institute (DSI) stepped in as sponsor. Organizers were Michael Ward and Amar Saini. Michael co-chaired the fall 2018 hackathon and was recruited by DSI director Mike Goldman to do the same this season. Adding a hackathon to the annual rotation may sound simple, but recreating the expected experience was tough. Michael points out, “Our goal was to make the winter hackathon look and feel like the others. I think we succeeded.”
3. Indeed, the event has become streamlined through repetition. The venue—Livermore’s High Performance Computing Innovation Center (HPCIC)—has two large, versatile conference rooms connected by a kitchen. Lessons learned include providing power strips at worktables and taking advantage of the building’s wheeled whiteboards.
Figure 1. Anh Quach and Lisa Her work to replicate a probability impact diagram with Tableau software.
4. DSI Council member Timo Bremer kicked off the hackathon with remarks about DSI’s role at the Lab. “The DSI was formed to coordinate data science activity internally and externally,” he stated. “It’s a resource for research questions, project information, and related events like reading groups.” Timo encouraged the audience to contact the DSI with suggestions for educational opportunities. Above all, he said, “The hackathon is a great event for getting people together. Have fun.”
5. Although hackathons typically do not have a theme, the DSI’s sponsorship inspired a sizable portion of participants’ projects. (Perhaps data science techniques also made their way into this hackathon because the field is growing at the Lab.) Michael notes that a future DSI-sponsored hackathon may introduce a data challenge for participants to tackle.
6. Rob Blake and Hardeep Sullan showed up to work on a machine learning (ML) solution for researchers who are new to ML. Their tool is designed to guide new users on ML best practices—processing data in Jupyter Notebook, selecting pre-trained ML models relevant for the problem, and training the models in parallel on Livermore’s parallel clusters. Rob, a hackathon regular, was inspired by the fatigue of “writing the same code over and over” whenever he wanted to spin up a new ML project. He explained, “We want to give researchers a starting advantage. Ideally, they could use our tool to create a working model in less than a day without having to learn the whole ML field. We managed to get a repository up and running on a Livermore Computing cluster, available for download by Lab users.” The team concentrated on image classification for this hackathon and hope to continue later with other well-defined data analysis problems, like time-series data.
7. In Marisa Torres’s nearly 18-year Lab career, this was her first hackathon. She is the organizer of Livermore’s Women in Data Science (WiDS) regional conference, which is coming up on March 2 alongside the global WiDS conference. Based out of Stanford University, WiDS holds a datathon competition in the weeks leading up to the conference. This year’s problem involves predictive analysis of patient data. “The winter hackathon coincided with the datathon timeline, so I thought this was a good opportunity,” said Marisa, who teamed up with fellow Girls Who Code volunteer Jessica Mauvais.
Figure 2. Matt Lambert describes his process of learning how to make a VTK (Visualization Toolkit) mesh data file that Livermore-developed VisIt visualization software can read.
8. Amar joined Luis Gutierrez once again to explore new uses for artificial intelligence. The pair embraces the event’s “try anything” spirit, this time experimenting with deep learning techniques to scrub noise from speech-based audio files. Amar explained, “We’re trying to convert audio into an image representation called a spectrogram, then distort the ‘bad’ parts. A neural network cleans the audio, and another one converts the picture back into audio.”
9. Inspired by applications of ML in medical imaging, one team used ML to label synthetic radiographs similar to those used for National Ignition Facility laser experiments. Kevin Chen, Charles Doutriaux, and Katie Lewis generated and labeled 2,000 synthetic radiographs to train a convolutional neural network that was originally created for identifying brain tumors. Katie noted, “In true hackathon fashion, the results didn’t completely make sense.” Nevertheless, the project provided an opportunity to transfer an ML solution to another scientific domain, and the team has ideas for next steps.
10. Everyone knows that differences abound between Mac and Windows operating systems. A straightforward task on one might be convoluted on the other, and sometimes developers go through different processes to accomplish the same results. Build systems are a prime example of this—a developer might use makefiles to compile a program on the Mac and Microsoft Visual Studio (VS) project files on a Windows machine. Arlie Capps noticed how quickly build systems for the same application can get out of sync. He used his hackathon time to write Windows-friendly build files using CMake, a tool to generate both makefiles and VS project files. To generate the kind of full-featured VS project files that Windows developers are used to, Arlie said, “You have to set CMake build parameters for each configuration, rather than for the one ‘current’ configuration.”
11. The web team behind the websites computing.llnl.gov, hpc.llnl.gov, and st.llnl.gov spent the hackathon period building an integrated Django dashboard to monitor server status, push database changes to production, and track other development environment metrics. Meg Epperly stated, “It’s nice to have dedicated time to learn new technologies and do team building.” Justin Mello added, “I’ve been wanting to learn Django for a while now and although I have dabbled with it in the past, it was really great to try to implement it with a project that provides value to our team. Working together in that capacity allows our team to think through problems and give input on how each person would tackle certain issues.” In addition to Meg and Justin, the team also included James Chalabi, Michaela Herman, and Kyle McAneney.
12. Computing staff clearly do not have hackathon fatigue. More than 60 participants gathered at the HPCIC over the two days, with many others working remotely from offices or smaller conference rooms. “I’m absolutely amazed at the turnout this soon after the holidays,” Michael said.
13. Shortstop Cal Ripken, Jr., may have played 2,632 consecutive games for the Baltimore Orioles, but he has not been to a Computing hackathon. For the 23rd straight time, the “big three”—Gary Laguna, Greg Lee, and Walt Nissen—were in attendance.
Figure 3. Stephen Herbein explains the work his team did to generate Rust bindings for Flux, a high performance computing resource management tool developed at the Laboratory.
14. Hackathon door prizes are not awarded by merely pulling a name out of a hat. A Python-based random number generator does the job, selecting from the number and order of participants who gave presentations at the event’s conclusion. Kyle Dickerson won this season’s prize: a lanyard stamped with the DSI logo.
15. The unsung heroes of hackathons are members of Computing’s administrative team. Each hackathon requires at least two people to manage venue reservations, setup, and cleanup; audio/visual equipment for project presentations; and, perhaps most importantly, the food. Among other tasks, administrators keep the snack table stocked and coordinate with caterers for lunch and dinner. Special thanks go to Jennifer Bellig, Brittany Robertson, and Tristan Jensen for ensuring the winter event ran smoothly and safely.
16. Last but not least: Laboratory photographer Randy Wong is not the sole chronicler of hackathons, but he has wielded the camera at 9 of the last 12 events, including this one.