SGN Retreat – October 14-18
Participants
- Chad Hanna (3,4)
- Yun-Jing Huang (4,3)
- Jim Kennington (3,2)
- Cort Posnansky (1,3)
- Patrick Godwin (3,4)
- Aaron Viets (1,2)
- Cody Messick (1,4)
- Surabhi Sachdev (1,4)
- Ryan Magee (1,4)
- Sarah Caudill (1,3)
- Leo Tsukada (1,4)
- Anushka Doke (1,1)
- Manvi Jain (1,1)
- Stefano Schmidt (1,4)
- Rhiannon Udall (1,1)
- Prathamesh Joshi (1,3)
- Ron Tapia (i,j)
- Victoria (1,2)
- Shomik (1,2)
- Zach Yarbrough (2,2)
- Shio Sakon (1,2)
Travel Info:
- LOCATION: 320 Whitmore (Institute for Gravitation and the Cosmos) University Park, PA
- For those who have requested a hotel, please work with Courtney asap on your reservation.
- Although sometimes a bit more expensive and with slightly less convenenient itineraries, it is almost aways the best thing to fly to State College Airport.
Tentative Schedule:
This is a working meeting. Bring your laptops. We will code nonstop from the first afternoon onward.
IT IS EXPECTED THAT EVERYONE WILL COMPLETE AT LEAST ONE MR PER DAY STARTING IN DAY 2 - PROBABLY MORE
Monday October 14: Quasi-circular
Time | Topics |
---|---|
09:00 - 09:15 | Introductions |
09:15 - 09:30 | Goals and motivations of this meeting |
09:30 - 10:00 | The path to parity |
10:00 - 11:00 | How to set up an SGN development environment |
11:00 - 12:30 | Your first SGN element and pipeline |
12:30 - 01:30 | Nourishment |
01:30 - 02:30 | Show and tell |
02:30 - 03:45 | Hands on with the CBC pipeline |
03:45 - 04:45 | The journey to transformation |
04:45 - 05:00 | The self you can give |
Tuesday October 15: Coalescing
Time | Topics |
---|---|
9:00 - 9:30 | Strategery |
9:30 - 12:00 | Hack session 1 |
12:00 - 1:15 | Provisions |
1:15 - 4:00 | Hack session 2 |
4:00 - 5:00 | Show and tell (you will be judged) |
Wednesday October 16: Merger
Time | Topics |
---|---|
9:00 - 9:30 | We are in this together |
9:30 - 12:00 | Hack session 3 |
12:00 - 1:15 | Sustenance |
1:15 - 4:00 | Hack session 4 |
4:00 - 5:00 | Show and tell (you will be judged) |
Thursday October 17: Disruption
Time | Topics |
---|---|
9:00 - 9:30 | There is no going back |
9:30 - 12:00 | Hack session 5 |
12:00 - 1:15 | Vittles |
1:15 - 4:00 | Hack session 6 |
4:00 - 5:00 | Show and tell (you will be judged) |
Friday October 18: Light
Time | Topics |
---|---|
9:00 - 9:30 | From stardust |
9:30 - 12:00 | Hack session 7 |
12:00 - 1:15 | Fare |
1:15 - 4:00 | Hack session 8 |
4:00 - 5:00 | Show and tell (you will be judged) |
Goals and motivations of this meeting
- Moderinize the offline (and online as time permits) GstLAL inspiral pipeline with SGN and PyTorch
- End up with no GstLAL dependency
The path to parity
- https://app.diagrams.net/#G1glg8R9ktXjPQO2jxlTvIRdI6JC9durlJ#%7B%22pageId%22%3A%22e9bE6tjlP6Qc1pHcuBe_%22%7D
- https://docs.google.com/document/d/1bSLZSaYFLcChQ6avb_57IzPL_2pxqF_RgXCj7zAjrNA/edit
How to set up an SGN development environment
- TBD notes from Yun-Jing and Jim (should be posted on actual documentation site)
Your first SGN element and pipeline
- JIM WILL DO THIS
Hand on with the CBC pipeline
- An example running on white noise (could be done on laptop)
- An example running on frame data (probably done on the cluster)
The journey to transformation
- Describe the plan of work for the week
The self you can give
- This is where people sign up for stuff to accomplish this week.
HACK SESSION ONE
-
Getting a complete replacement for the offline PSD DAG
- Write sgn_reference_psd (Ryan Magee, Anushka Doke, Shio)
- Write sgn_median_of_psds (a cut and paste from gstlal with care given to how it fits into the libraries) (Ryan Magee, Anushka, Shio)
- Write an sgn_reference_psd_dag built on ezdag OSDF aware (Cort Posnansky, Patrick Godwin)
-
Porting and improving the SVD code
- Make an sgn_inspiral_svd_bank (mostly a cut and paste) (Surabhi Sachdev, Manvi Jain, Stefano Schmidt)
-
Getting a complete injection frame dag working
- SGN LIGO injection source - CBC only for now (Victoria Niu, Jim, Sarah)
- Frame Writer (Zach, Shomik)
-
Strike dev list
- strike_inspiral_create_prior_diststats (Leo Tsukada, Prathamesh)
- Revisit SNR-chisq histogram object see https://git.ligo.org/greg/greg/-/blob/main/greg/types/cereal.py?ref_type=heads#L52
- Ditch xml format and Implement to_h5/from_h5
- Clean up/rework far.py
- Speed up train_noise function (Yun-Jing doesn’t like it)
- Profile train_noise function
- Implement masked SNR-chisq sampling https://git.ligo.org/lscsoft/gstlal/-/commit/9f36f8fa5f14ceb4200a8b6909d6d9864feeb07f
- Implement masked dtdphi sampling (This might not work with the masked SNR-chisq sampling, i.e., too much masking) We have never done this in all sky dtdphi yet, but we can follow this commit https://git.ligo.org/lscsoft/gstlal/-/commit/eddbf1612c4520f1eaf32a0a8712a4f29a29a45e
-
Output files
- Hardening output file stuff - add segment tables, process tables and the like? (Yun-Jing Huang, Chad Hanna, Jim Kennington)
-
Handling runtime control
- Bottle (Ron, Yun-Jing, Chad, Jim)
- Signal Handler (Ron, Yun-Jing, Chad, Jim)
HACK SESSION TWO
- Make an sgn_inspiral_svd_bank_dag built on ezdag OSDF aware (Cort Posnansky, Patrick Godwin)
- [SVD bank conversion to H5 (Ryan Magee)]
- SGN LIGO injection source - CBC only for now (Victoria Niu)
- [strike_inspiral_marginalize_likelihood (Leo Tsukada)]
- Combine triggers across times / cluster triggers by XX / Simplify triggers??? / merge and reduce (Cody Messick, Ryan Magee, Surabhi Sachdev)
HACK SESSION THREE
- Calc likelihood (???)
- Extinct bin (???)
- Calc rank pdfs (???)
- Compute far (???)
- Trigger dag with ezdag and OSDF aware
- Plots, plots, plots
- The injection workflow (formats, splitters, snr, injection finding)
HACK SESSION FOUR
- Rank dag with ezdag and OSDF aware
- Plots, plots, plots
- The injection workflow (formats, splitters, snr, injection finding)
ACTIONS BEFORE THE MEETING
- “finish” stillsuit / sgn-event integration
- Standardize on stride offset? get some elements converted over?
- LIVE Test source
- Code quality control in place for all the sgn*, stillsuit, etc. - Get it all development ready.
- Getting dev environments working trivial.
- Demo where people can download frame files from GWOSC for the CBC pipeline.
NOTES
- bottle implemented as a source element and elements that can be controlled have flow control sink pads.
RANDOM STRETCH GOALS
- NOMAD?? https://www.nomadproject.io/