Main_ipac_logo

March 23 - 27, 2026 • Hameetman Auditorium, Caltech

Hack Session

We have organized five hacking tracks based on the RAPID pipeline processing of simulated Roman data. Please download the RAPID prototype data products, and the starter notebooks in the tar file here: https://caltech.app.box.com/s/a853szhuniu3on2rgwxj99trcbmf69j0 and select one track.

If you have other ideas for a hack track, please let us know and we will be happy to discuss options. Please propose by the end of Wednesday, March 25.

Why bother with the hackathon?

We are rapidly approaching Roman's launch (no earlier than Sept 2026)! After only a 90-day commissioning period, RAPID data products will be available to the community. Get ready with this hackathon for some fun infrared time-domain science! Also, don't miss the little prizes that will be awarded on Friday after the presentations by the hackathon track teams.

What are the Five Hacking Tracks?

  • Alert brokering: filtering, ranking, classification (TA Matthew Graham, mjg AT caltech.edu; #hack-session-track1)
    • Challenges: We provide the ingestion code for RAPID alerts. Can you improve filtering to identify more transients? Can you incorporate methods that use fewer light curve points, or varied filters for further subclassification? Can you build a quality filter and rank candidates?
       
  • Source detection in difference images (TA Jacob Jencson; jjencson AT ipac.caltech.edu #hack-session-track2)
    • Challenges: Can you optimize source detection on RAPID difference images to find the most reals? Can you optimize the code to reduce false positives? Can you code to use alternate methods for better detection?
       
  • Real/bogus classification (TA Ashish Mahabal, aam AT astro.caltech.edu; #hack-session-track3)
    • Challenges: Can you optimize your ROC score when finding reals amidst the fog of bogus sources in the RAPID alert stream? Can you cluster bogus events so that information there can be used to improve real-bogus separation? Can you come up with image-based features that improve the separation? Can you incorporate active learning?
       
  • Star/galaxy separation (TA Ryan Lau and Ilaria Caiazzo, ryanlau AT ipac.caltech.edu and ilaria.caiazzo AT ist.ac.at; #hack-session-track4)
    • Challenges: Can you use the stacked RAPID reference image to robustly differentiate between a star and a galaxy? Can you come up with image features that separate the two classes better?
       
  • Light-curve analysis and transient typing (TA Wynn Jacobson-Galan, wynnjg AT caltech.edu; #hack-session-track5)
    • Challenges: Can you use the RAPID light curves to find supernovae? Can you find the variable stars? Can you use fewer points in the light curve and still figure out the classification? Can you do multi-modal fusion for better classification?

What is included in starter notebooks?

  • Start with the 00_data_overview notebook
  • Create virtual environment: conda create -n rapid python=3.12
  • Activate environment and install requirements: pip install -r alerts/notebooks/requirements.txt
  • See the data access and loading instructions
  • See the suggested problem-solving directions

If you are stuck, ask your TA for hints/approaches or even the baseline, non-optimized, ML solution notebooks!

What is RAPID and what are the RAPID data products?

RAPID is a Project Infrastructure Team for the Roman Space Telescope. All RAPID development is open source and you can find all the code details and latest, extended datasets and corresponding read the docs here.

We have selected a small subset to play with in the hackathon and provided that at the box link above for you to download. If you have trouble downloading, ask your TAs for a USB stick to get the data.

When are the hack sessions in the workshop schedule?

Thursday, March 26 (15:30 - 17:00)

  • 15 min: overview + team formation - Hameetman Auditorium, Cahill Building
  • 75 min: hacking session, Hameetman + Keith Spalding Building, Fourth Floor

Friday, March 27 (13:55 - 15:25)

  • 60 min: Continued hacking, Hameetman + Keith Spalding Building, Fourth Floor
  • 25 min: Short presentations per Hackathon Track, Hameetman Auditorium, Cahill
  • 5 min: Prize distribution and next steps, Hameetman Auditorium, Cahill Building

After the workshop

Participants are encouraged to improve on their notebooks after the workshop. We will host special STRIDE meetings (April 3, 2026 and May 1, 2026) for you to present your final results. We will also summarize the hackathon results at the Roman quarterly meeting (April 14–16). Participants are encouraged to start early and self-organize into teams.