Tuesday, 3pm - 4:30pm CST
Attendees: Bill Tang, Shinjae Yoo, Yi Huang, Alexey Svyatkovskiy, Kyle Gerard Felker, Jesse Rodriguez, Felipe Gomez
Link to Box folder with slides and other files:
- Needs to reactivate his Princeton Research Computing account;
- Contact Eliot for help converting the account to an external collaborator status and reactivating it?
- Needs to copy fork of PyTorch
timm
- Currently reproducing Michael Churchill's 2019 NeurIPS workshop paper on ECEi data with
/scratch/gpfs/rmc2/ecei_d3d/data/- This data is still at the original 1 MHz sampling. He downsampled on-the-fly
- Transferring the 10 TB to Yi's local workstation; may take 2 days. Local workstation only has 1x GPU (32 GB V100?), so won't do data parallel training at first
- ~2K shots in training set used in
- We have at least 6K shots with both LFS and HFS channel data from ECEi. Should expand Churchill's 2019 database when reproducing
- ~25% of the shots have significant low-frequency noise and the channels appear "dead" (std dev < 1 mV); this is different than the significantly greater number of shots will some or all channels "missing"
- Globus DTN at Princeton is limited to 500 GB at a time for Jesse's account. No such limit on ALCF DTN? We can possibly use that as the centralized store for sharing with Brookhaven, etc
Locations of ECEi data on Princeton Tigress:
/projects/FRNN/signal_data_new_REAL_TIME/d3d/ECEI_100kHz//projects/FRNN/signal_data_new_REAL_TIME/d3d/ECEI_HFS_100kHz/
On Polaris: /eagle/fusiondl_aesp/signal_data/d3d/ECEI_100kHz
-
Seeking input: Fz collaboration for data compression needs for our application
-
Reactivate Alexey's ALCF account for Polaris access
-
Schedule next meeting, Jan 3rd or 4th probably (Shinjae conflicted first two weeks of January; Yi can likely make it)
-
Additionally, Alexey to sync wtih Kyle, Felipe, Jesse after new year in 1-1 meeting(s)?