Difference between revisions of "BMOD 20Sept2021"
From PREX Wiki
m (→Agenda) |
m (→Sept 20) |
||
Line 14: | Line 14: | ||
* To do: | * To do: | ||
** Double check smoking gun causes of outliers multiplet and minirunwise | ** Double check smoking gun causes of outliers multiplet and minirunwise | ||
− | + | ** Throw away miniruns of bad problems (7211) | |
Revision as of 08:13, 20 September 2021
BMOD_17Sept2021 << >> BMOD_21Sept2021
Contents
Logistic information
BlueJeans calling instructions: Toll-Free Number (U.S.& Canada): 888-240-2560 International toll number: 408-740-7256 Bluejeans CODE: 948942477 Bluejeans link: https://bluejeans.com/948942477
Agenda
Sept 20
- To do:
- Double check smoking gun causes of outliers multiplet and minirunwise
- Throw away miniruns of bad problems (7211)
- Standard set of plots from before
- Multipletwise "mul plots"
- Log scale, gaussian fit, us_avg
- Log scale, gaussian fit, us_dd
- Regression check - Log scale, gaussian fit, us_avg (regression and Lagrange are similar!)
- Linear scale, zoom in on the central peak, us_avg
- Check all currents, not just >135 uA this time
- Outliers checked
- minirunwise pull plot outliers
- multiplet plot RMS outliers
- greater than 10000 ppm outliers. Come from Flip LEFT state, but are distributed sporadically throughout the span from slug 136-211.
- of the 23 outlier runs, 4 show behaviour in the raw data that might explain the outliers. All other runs look good.
- Minirunwise fit and pull plots
- Kinematics variables considered (target X, Y, theta X, theta Y, arc dispersion E)
- All in one PDF (page 2)
- And also along with the rest of the plots directly below:
- Various time scale averaging of quantities of interest, and table outputs
- Table outputs of monitors and corrections in one place
- (sorry, I haven't given them the excel treatment or the above table's latex output yet, but I plan to make another script for it)
- NULL Pitt
- CREX "Parts"
- 6 IHWP x Wien states (Wien 3 has substantial cancellation between IHWP IN and OUT!)
- 3 Wien states (Wien 2 gets 2% of CREX part 1's data, listed below, which makes the nice 0 correction bump up to 3ppb correction now)
- Changing from "part" averaging to "wien" averaging, these are the runs that are CREX part 2 but belong to wien state 1, i.e. the 1 day's worth of running in a new optics set up in the old wien state before going to AT running:
- Minirun count through the experiment (total are 8798), CREX "part", run number, Wien flip number
- 1316 , 2 , 6328 , 1
- 1317 , 2 , 6328 , 1
- 1318 , 2 , 6328 , 1
- 1319 , 2 , 6328 , 1
- 1320 , 2 , 6328 , 1
- 1321 , 2 , 6328 , 1
- 1322 , 2 , 6329 , 1
- 1323 , 2 , 6329 , 1
- 1324 , 2 , 6329 , 1
- 1325 , 2 , 6329 , 1
- 1326 , 2 , 6329 , 1
- 1327 , 2 , 6330 , 1
- 1328 , 2 , 6330 , 1
- 1329 , 2 , 6330 , 1
- 1330 , 2 , 6330 , 1
- 1331 , 2 , 6330 , 1
- 1332 , 2 , 6331 , 1
- 1333 , 2 , 6331 , 1
- 1334 , 2 , 6331 , 1
- 1335 , 2 , 6331 , 1
- 1336 , 2 , 6331 , 1
- 1337 , 2 , 6331 , 1
- 1338 , 2 , 6331 , 1
- 1339 , 2 , 6332 , 1
- 1340 , 2 , 6332 , 1
- 1341 , 2 , 6332 , 1
- 1342 , 2 , 6332 , 1
- 1343 , 2 , 6332 , 1
- 1344 , 2 , 6332 , 1
- 1345 , 2 , 6333 , 1
- 1346 , 2 , 6334 , 1
Sept 17
- Multiplet plots (multiple aspect ratio examples), and IncludeBMOD lagr asym and ErrorFlag kinematics variables examples
- All single plots examples
- Previous standard averaging plots
To look at:
- Lagr asym main det, IncludeBMOD Lagr asym main det, Lagr asym DD, IncludeBMOD Lagr asym main det
- BPMs vs. parts table - table example
- evMons vs. parts table
- Kinematics BPMs vs. parts table
- Multiplet plot of Lagr asym main det > 135 uA - Rootfile lives here: /lustre19/expphy/volatile/halla/parity/crex-respin2/pruned_lagr_analysis/All_Lagr_Production.root
- Multiplet plot of Lagr asym DD > 135 uA
What to plot next:
- Calculate per time-step (parts, slugs, etc.) the maindet-weighted average mean value. Error bar (in the TGraphErrors) should be minirun-mean_error weighted by detector stat-weight (mean_err, plot the mean_err itself, no rescale!) -> but don't fit that on the TGraph, do calculate the "global" average value+err and show it in statbox/overlay.
- Null Pitt - The In and Out states should each have 1/2 weight into the average.
- Do Wien's not crex_parts (double check slow_controls output, and add a 1,2,3 Wien variable).
- Check NULL pdf works
- Get BCM, BPM tables, pdfs fast
- Send data and scripts to Weibin
- Check minirunwise pull and multipletwise mul_plot outliers
- Make slugwise plots (add a main-det weighted mean+err to statbox, in addition to self weighted p0 fit value... no fit line at all)
Sept 10
https://docs.google.com/document/d/1tnJQ0Ic679W8Gxl1VNAS5D2oNhxCgXl0kK_V9ee4Ch8/edit
- Correction certainty stress test: Giant Spike y-rms runs (7500-7715, slugs 185-187 ish)
- Corrections (total and per evMon or bpm) agree between methods?
- Cameron slug level correction agreement type up log
- Minirunwise correction agreement (write a aliasing loop for it)
- Explanation of regression’s limitations: Generally large regression RMS 3rd period
- Y mod only? Or…?
- Y corrections dominate regression discrepancy?
- Bmodonly, reg, coil1,3,5,7 => if good, then not X’s fault
- Bomdonly , reg, coil2,4,6 => if bad, then Y’s fault
- Sanity check of BPM independence: diff_bpm12X_RMS - y mod only, still bigger?
- Curiosity: Why are OnlyBMOD X and Y BPM RMSs doing weird things? Quartet synching?
- e.g. X-rms seg 2, phase of modulation?
- Plot of: yield_beam_mod_ramp_mean and _rms vs. diff_bpm4eY_rms per minirun. If _mean or _rms are correlated the the BPM's RMSs at all this explains it. The 8-step phased bmod cycle can be 2 multiplets of phase 1 and 5, 2 and 6, 3 and 7, or 4 and 8. In yield_beam_mod_ramp units this appears as the following plot:
- e.g. X-rms seg 2, phase of modulation?
- The fact that the starting phase (or average phase, however it is calculated) of the multiplets within supercycles is unstable to span multiple starting points of phase options is a sign that there is phase drift and a plot as described before (phase vs. bpm rms) would be useful.
Sept 8
BMOD Inclusion
Plan
Check all of these minirunwise and slugwise, for ErrorFlag, IncludeBMOD, and OnlyBMOD compared to eachother
- Check corrected RMS difference between all 3 cuts for the analysis techniques
- 12BPM Eigen Lagr
- 12BPM Eigen Reg
- 5BPM Eigen Dit
- Plots:
- Corrected RMS for each cut, for each analysis technique
- Corrected RMS difference between cuts difference, for each analysis technique
- Corrected RMS ratio between cuts difference, for each analysis technique
- Check the mean value corrected asymmetry difference (minirunwise differences) between all 3 cuts, and pull plot
- Plots:
- Asymmetry difference between cuts p0 fit, for each analysis technique
- Asymmetry difference between cuts pull plots, for each analysis technique
- Plots:
- Check the mean value corrected asymmetry difference (multipletwise if possible) of each 3 cuts between analysis techniques
- Mean value difference: 12BPM Eigen Lagr vs. 12BPM Eigen Reg vs. 5BPM Eigen Dit, for each cut
- RMS of that difference: 12BPM Eigen Lagr vs. 12BPM Eigen Reg vs. 5BPM Eigen Dit, for each cut
- Check the BPM differences means and RMSs, check if any unreasonable noise is introduced, between all 3 cuts
General minirunwise grand plots (using self errors for weights everywhere)
- Allbpms, ErrorFlag cut - pdf, txt
- Allbpms, IncludeBMOD cut - pdf, txt
- Allbpms, OnlyBMOD cut - pdf, txt
Specific minirunwise comparison plots
- Include BMOD vs. ErrorFlag - with the really bad outlier removed
- OnlyBMOD vs. ErrorFlag - with the really bad outlier removed
Sept 1
Next steps and priorities:
To finalize the beam corrections results:
- Make tables of the position differences and evMon diffs (separate the 3 parts)
- Augment all tables and plots, separating the 6 IHWP x Wien states out
- Consider the target position and angle and E BPM definitions (if possible)
- Make a Lagr mul plot (all multiplets in 1 histogram)
- Archive plots, tables, and results of slug, pitt, etc. plots in a useful format, documented, publicly (haplog/ifarm tape/docdb)
- Do NULL Pitt plots as well
- Treat Main det double difference alongside A_PV, corrections, multipletwise differences, etc.
- Make a multipletwise difference distribution RMS plot (minirunwise, slug, etc.) in addition to the mean disagreement numbers
To determine whether to include BMOD in the final dataset do:
- Produce aggregator data for Lagr-12BPM dataset with Include and Only BMOD cuts
- Look at the corrected asym RMS, separately with OnlyBMOD and ErrorFlag cuts
- Check if the IncludeBMOD corrected asym mean and correction per monitor means jump around outside of statistical fluctuations, compared to ErrorFlag cut dataset
- Check the multipletwise difference between analysis methods means and RMSs, if OnlyBMOD is substantially worse than ErrorFlag cut dataset
Document, commit, etc. all necessary scripts, and make sure data is saved in safe places
August 25
Respin 2 beam corrections results - slides
August 18
Plan:
- Data quality check: Minirun and Slug-wise check outliers (us_avg, usl, usr together) of
- Check 5bpm dit vs. 12bpm Lagr (page 1) and 12bpm reg vs. 5bpm dit (page 9) - Check comparisons of USL vs. USR and pull plots...
- Method-differences plots
- Corrections per monitor
- Method slopes differences (should be benign anyway) (page 17 lagr vs. reg, etc.) - scan
- Split CREX into the 3 parts -> look for outliers from means that are ~2 sigma pull (scan the tree in part-averaging script, and plot the data in minirun-wise plot) and evaluate potential nefarious causes in the run-wise raw data (usl, usr, us_avg)
- Lagr vs. reg slope comparison (allbpm eigenvector analysis)
- Comparing correction per monitor also between Lagr vs. reg (prior times diff_evMon_mean)
- Corrected asyms differences also between Lagr vs. reg
- Pitt and slow control averaging plots (signed and unsigned)
- Evaluate including BMOD in the final-polish/verified data-set (do slug avg, compare means and RMSs of corrected asyms between dataset)
July 21
Looking at results from respin 2 data:
- dit - reg - lagr differences
- impact of including error bars on lagrange sensitivities
- slope plots
- slope differences per monitor from different techniques (dit vs reg)
- accumulated residuals (or, residual spread per slug RMS and averages)
- net beam corrections (3 parts and slugwise)
Tasks for next week:
- Try to zoom in on the parts where plain and eigen dit corrected asyms disagree
- Double check the root files and slopes, etc. that there is nothing wrong in the analysis chain
- The two methods should get identical corrected asyms
- Verify no 1 arm running shenanigans or mislabeled data slipped in (and that run_avg == eigen_dit_run_avg_part_avgd)
- Look at where slopes disagree between dit - reg
- Verify the corrections and/or monitorwise diff RMSs are small (so that if the slopes disagree it has negligible effect on the corrections being made)
- Make slug_avg plots of the slope disagreements (including RMSs)
- Make slug_avg plots of the corrections per monitor too
- Do a 12BPM Eigen Reg and Eigen Lagr comparison
- See if the slopes agree better than the 5BPM case (where evMons 3 and 4 disagreed substantially between reg and dit)
- Provide easy to parse tables of the 5 evMon's corrections per 3 CREX parts (and make the error calculation math clear as well)
- Improve dit-reg-lagr plots
- Add pull plots
- Do slug averaging of these (including RMSs)
- Multipletwise and runwise diff plots being == is suspicious
- Double check the plotting script isn't doing anything funny
- Double check the source aggregator root file and input device list are correct
- Double check the math by plotting the distributions by hand for some example runs
July 14
- Respin 2 outputs
- Updated eigenvector definitions
- Dit + reg + lagrange using eigenvectors
- Residual sensitivities, done correctly this time
May 19
- Respin 2: Look at dit vs. reg differences - HAPLOG
- Look at respin2 dit-reg diff outputs
- Checked by eye the 4+ sigma outlier miniruns
- Respin 1: 3 CREX parts’ monitorwise net corrections (and uncertainties?) - HAPLOG
- Respin 1: Residual dithering sensitivities - HAPLOG
- Looking at plain BPMs, run_avg sensitivities
- Looking at eigenvector monitors, run_avg sensitivities
- Looking at plain BPMs, cyclewise sensitivities
- Fractional residuals (needs more work in respin2 outputs to clean up outliers) - HAPLOG
- Timescale dependent RMSs in asym->histogram filling - not discussed - HAPLOG