1 / 27

Where are we ?

Photonics Tables Bin Optimization Kyle Mandli Paolo Desiati University of Wisconsin – Madison Wuppertal AMANDA Collaboration Meeting. Where are we ?. Stephan: comparison between PTD bulk tables and Photonics bulk tables (muon, shower)

Download Presentation

Where are we ?

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Photonics TablesBin OptimizationKyle MandliPaolo DesiatiUniversity of Wisconsin – MadisonWuppertal AMANDA Collaboration Meeting

  2. Where are we ? • Stephan: comparison between PTD bulk tables and Photonics bulk tables (muon, shower) • Check if Photonics bulk tables are consistent with PTD • New AMASIM release and bulk tables test • Thomas: implement PSI interface for Photonics, PTD and NN-fit tables • Johan: work with Thomas in PSI and produce Stephan’s test using PSI • Check results consistency with different interface (IceCube) • Daan: working on NN-fit procedure of Photonics tables and produce comparison tests with tables themselves • Check if NN fit are a good approximation to speedup simulations

  3. Where are we ? • Ignacio: implement the zenith bin-wise production • Another simulation speed up possibility • Adam,David H.: check memory map feasibility • Yet another simulation speed up possibility • Kyle M.: Photonics tables bin optimization • Come up with a binning as a good compromise between good ice description and an acceptable table size Bin optimization is the topic of this talk

  4. 1-slide tutorial • Light tracking binning: • Photons tracked using 6 parameters: ρ, φ , z, t, θe, θa • This affects the table size • Light source binning: • depth (z) & angle (θ) binning • This affects the number of tables • Table types: • Point-like ems and muon tables (differential or level1 tables) • Infinite muon tables (extended source or level2 tables) • Time-integrated amplitude PDF tables (.abs, smaller tables) • Integral PDF time tables (.prob, full binning tables) • Differential time probability table http://amanda.wisc.edu/simulation/photoproduction/tables

  5. Bin optimization proposed procedure • Produce single tables @ given source locations (z=0): only ice properties • Produce this table with different tracking binning in ρ, φ, z, t, (θe,θa integrated) • with statistical errors (interfaces do not read errors at the moment)

  6. Optimization procedure • Use Time delay distribution @ given source-receiver dist d • Receiver at reference origin • Light sources (10 GeV ems) on z=0, random in circle at a given d • Sampling different table projections as in a simulation • Mean Amplitude & Mean Time Delay versus distance d • Light sources on z=0, random in disk up to ρmax • Perform a statistical test on ems tables • Kolmogorov-Smirnov test between most dense table and the others • Calculate max y distance of cumulative histo and the probability that the 2 histograms are generated by a random sampling of the same distribution • Use of PSI => advantage of root => still in development (but new release now)

  7. Time delay distribution 20,000 random samples

  8. Time delay distribution 20,000 random samples

  9. Time delay distribution 20,000 random samples

  10. K-S Test: what we expect • Compare the most dense table with the others • Calculate the probability that tables in each pair is derived by random sampling of the same distribution • Test statistic depends on K-S max distance and number of entries in each histo Prob Increase prob tab-pairs Increase bins

  11. K-S Test

  12. K-S Test

  13. Table bins reminder • Tab 5 seems to be ~10% consistent with the most dense table • Statistical errors from tables (not accessible) and MC simulation

  14. Mean Amplitude & Time Delay vs distance

  15. Mean Amplitude & Time Delay vs distance

  16. Mean Amplitude & Time Delay vs distance

  17. K-S Test

  18. K-S Test

  19. Table bins reminder • Given statistical fluctuations and table size: Tab 5 good compromise

  20. Other tests • A set of tables where only one dimension tracking binning is varied • The test suggests that changing binning in one dimension does not significantly affect the tables precision for a relatively wide range of binning. • A set of infinite muon tables to perform the same test • This test was NOT done so far • Would require other tables … • The source location binning to be tested (= layers) • NEXT STEP : it requires the generation of whole set of tables

  21. Conclusion on tracking binning • Suggested table with enough precision and reasonable size • We already used this. Not much improvement ! • Infinite muon Tables • These tables have not been tested (educated guess so far)

  22. Source location binning • What we used so far • With these values we have • 779 ems.prob + 779 ems.abs = 1558 ems tables = 2.34 GB • 779 mu.prob + 779 mu.abs = 1558 mu tables = 4.67 GB • 779 mu.prob + 779 mu.abs = 1558 mu tables = 2.34 GB • Total size : ~ 7 GB (4.67 GB)

  23. Size and Speed • Angular splitting in simulation • Load only the tables corresponding to all z-values and to only the 2 θ-values around the muon track zenith angle • Gain a factor 18/2=9 table size to load = 520 MB • Use of NN fits of the tables • Do not load any table but use the fit function • The function is complicated but <<< 1GB • Speed seems to be very competitive versus tables • Precision under extensive check • Memory mapping • Load tables (or portion of tables) on disk and access them using specific algorithm • Under investigation. More complicated than it seems.

  24. Final Table Production • Produce baseline tables (ems, muon diff tables) • Photon survival probability only with ice properties • Binned in θa • Store those tables : 86 GB but only once ! • Different efficiencies can be included without re-generating tables • Include efficiencies, and produce .prob, .abs and .diff • Special tables: • UHE, monopole tables : wider ranges (up to 1000 m in z and ρ) • .diff (dP/dt) tables for reconstruction with finer bins and less dimensions • Propose to generate them separately

  25. Final Table Production • Efficiencies to be included • Like in previous tables production ?

  26. Conclusions • Lots of progress recently and still on the way • Waiting for layered tables: will be on disk next week • Still waiting for me ? I remind that we already have tables ! • Efficiencies can be changed faster • Start simulation for further tests • Purely interface reading (i.e. PSI) • AMASIM runs: speed • Theta angle binning of muons • NN testing • NEED OF PEOPLE CURIOUS ABOUT THE SECRETS OF PHOTONICS

  27. Conclusions http://amanda.wisc.edu/simulation/photoproduction/tables

More Related