|
|
| (3 intermediate revisions by the same user not shown) |
| Line 24: |
Line 24: |
| | * Looks good so far | | * Looks good so far |
| | * Fixed an issue with bmod slopes (had the wrong sign for correction) | | * Fixed an issue with bmod slopes (had the wrong sign for correction) |
| − |
| |
| − | === New issues noticed: ===
| |
| − | There are a few runs that are somewhat orphaned, taken between major accelerator or magnet problems and which are hard to assign a trustworthy dithering slope. That isn't necessarily a big concern, but these same runs also appear to totally blow up the eigenvalue regression slopes (which shouldn't really be related to poorly selected BMOD slopes...). It may take me a while to figure out what is going on with these runs, and it could be some kind of issue with the septum magnet at that time that didn't get properly flagged in the RCDB.
| |
| − | * Looking at beam corrections, it appears that something fatal happened between runs 7743 and 7779. There are 3 production runs that are much worse in dithering than regression (7759, 7776, 7778), and are absolutely atrocious using the eigenvector regression (which is a big red flag). The wac_comment on RCDB for these runs indicates that bad things were happening around this time... I think we were down due to accelerator and Hall A magnet problems for 99% of 3 days at the time... so I think these runs may be suspicious
| |
| − | ** The annoying thing is that regular regression seems to do fine across these runs, but the choice of eigenvectors is clearly wrong and the choice of BMOD slopes is also somewhat wrong. Unfortunately the issue is that there is no good data nearby these runs from which to chose the BMOD slopes, and I guess the eigenvector regression broke for some worse reason...
| |
| − | ** Probably if BPMs are too correlated the eigenvector regression would fail. But the thing is that I averaged the eigenvectors over a long time scale, so it isn't the actual vector definition that is bad, but something about the beam conditions during these few runs that makes the vectors I chose to be very bad, I guess. These few runs are also the ones where dithering fails to make a correction as good as regression the most over the whole run. Just the eigenvector version totally gives up...
| |
| − | file: root respin1_data/aggregator_outputs_prod_only_mini_friendable.root
| |
| − |
| |
| − | draw eigen reg: agg_part_avgd_friendable->Scan("rcdb_run_flag:rcdb_run_type:asym_us_avg_eigen_reg_reg_diff_mean/1e-6:asym_us_avg_eigen_reg_reg_diff_mean_error/1e-6:rcdb_slug:run_number:minirun_n:rcdb_run_flag:rcdb_run_type:dit_asym_us_avg_mean/1e-6:dit_asym_us_avg_mean_error/1e-6:reg_asym_us_avg_mean/1e-6:reg_asym_us_avg_mean_error/1e-6:eigen_reg_asym_us_avg_mean/1e-6:eigen_reg_asym_us_avg_mean_error/1e-6","rcdb_arm_flag == 0 && (((rcdb_slug==190 || rcdb_slug==192) && (run_number>7743 && run_number<7780)) || (abs(reg_asym_us_avg_mean_error - eigen_reg_asym_us_avg_mean_error)/1e-6 > 0.25)) && rcdb_run_type==1 && rcdb_run_flag==1")
| |
| − |
| |
| − | draw plain reg: agg_plain_friendable->Scan("rcdb_run_flag:rcdb_run_type:asym_us_avg_eigen_reg_reg_diff_mean/1e-6:asym_us_avg_eigen_reg_reg_diff_mean_error/1e-6:rcdb_slug:run_number:minirun_n:rcdb_run_flag:rcdb_run_type:dit_asym_us_avg_mean/1e-6:dit_asym_us_avg_mean_error/1e-6:reg_asym_us_avg_mean/1e-6:reg_asym_us_avg_mean_error/1e-6:eigen_reg_asym_us_avg_mean/1e-6:eigen_reg_asym_us_avg_mean_error/1e-6","rcdb_arm_flag == 0 && (((rcdb_slug==190 || rcdb_slug==192) && (run_number>7743 && run_number<7780)) || (abs(reg_asym_us_avg_mean_error - dit_asym_us_avg_mean_error)/1e-6 > 0.25)) && rcdb_run_type==1 && rcdb_run_flag==1 && minirun_n == 0")
| |
| − |
| |
| − | After much soul searching I have learned many things about the data that I myself made.
| |
| − | Document things
| |
| − | * The aggregated numbers I had were actually generated using a long-time period averaging to obtain stable slopes, not doing the burst-by-burst slugs that I thought I had done (it's 2 months ago... I forgot)
| |
| − | ** So, the crazy large numbers that are outliers -> Only outliers because I was accidentally including 1 arm only data in the calculation, hence why only the right arm slopes got so skewed
| |
| − | ** This would have not been a problem if I'd properly segmented that data by arm flag previously... but I'll fix that now I guess
| |
| − | * The analyzer I got from Tao was in fact NOT giving the wrong answers
| |
| − | ** It has always given the right answers (confirmed now by just multiplying the burstwise slopes that should have been run, rather than the mixed arm-flag segment averaged slopes that were poluted)
| |
| − | ** It just turns out that regression will always give you the same final regressed asymmetry, no matter how you mix up the BPMs involved in the calculation
| |
| − | ** Which I guess is obvious, but I had been confusing myself with the time period averaging of slugs that gave slightly different answers than just burstwise regression
| |
| − | * Document things better
| |
| − |
| |
| − | So now I just need to ignore a few of the outputs I had, which include the polluted 1 arm running slope info (the dedicated comparison of different regression and dithering distributions to eachother check and the eigen_reg aggregator quantities...), and not use the segment averaged slope information that I mistakenly put in there for the special eigenvector regression analysis (I don't remember wanting this... but apparently I did at some point). Japan trees in respin2 will save a bunch of hassle on these things, but not enough I guess. This does mean we will need to rerun respin2 on slugs 190, 191, and 192, whose BMOD slopes were calculated with the 1 arm running pollution (I think... gonna check that one tomorrow). (edited)
| |
| − | 
| |
| | | | |
| | == March 30th 2021 == | | == March 30th 2021 == |
PREX Main << Weekly Analysis Coordinator << WAC Notes CREX << WAC Post-Run_Meeting Notes
WAC Notes March 30 2021 << >> WAC Notes April 27 2021
Organization
Date and time: April 20, 2021, 11am
Bluejeans ID: 564945377
Meeting link: https://bluejeans.com/564945377
Runlist spreadsheet
April 20th 2021
Respin 2 is going
- Looks good so far
- Fixed an issue with bmod slopes (had the wrong sign for correction)
March 30th 2021
- BCM and BPM Pedestals get a once over: http://ace.phys.virginia.edu/HAPPEX/4507
- Ifarm prompt-install is set up for respin2 (almost finished - still need to pull crex-respin2-alpha and recompile JAPAN).
- Run lists updated - still need to cross check with python sql database interacting scripts
- Other items from before are all addressed (?)
New RCDB/runlist updates
- Cameron has added the label "Good" to a bunch of worthwhile calibration and pedestal runs that need to be kept around for the future (listed at the bottom)
- Cameron has changed some runs to improve our coverage of nearly-production conditions and remove "Test" from our set of respin2 runs
- Change 5907 Test NeedCut to Production Suspicious (it is a slow target burn in run on Ca48 at low current).
- Change 6115 from Test Suspicious to Production Suspicious
- Change 6525 from Test Good to Production Bad (as there is some fatal blinder/target move issue and the run is very short and has messed up FFB/coil7 response)
- Change 8367, 8388, and 8389 from Test Good to Junk Suspicious (Carbon beam off-ramp ups)
- Change 6038-6043 from Test Good to Junk Good (DAQ testing runs)
- Cameron and Weibin triple checked 2 C++ scripts and a python script, after cleaning up the rcdb a bit more and utilize the following cut:
run_flag = NeedCut,Suspicious,Good, run_type = Calibration,Pedestal,Parityscan,Production and not cut on target
- This yields 1583 total runs, including some that are "Home" target (pedestal scans, etc.)
- 1404 are CREX Production (so 179 are Calibration, Pedestal and Parityscan) with 56 runs being AT only
- 1392 are CREX Production Good
- Python version /u/group/halla/parity/software/japan_offline/prompt/prex-prompt/WAC/auto_run_list/crex-respin2/CREX-All-run-list.list
- C++ version /u/group/halla/parity/software/japan_offline/prompt/prex-prompt/prex-runlist/crex-respin2/all_nonJunk.list
March 23rd and before
WAC Notes March 23 2021
Attendence
