Skip to content

Latest commit

 

History

History
885 lines (675 loc) · 24.7 KB

HISTORY.rst

File metadata and controls

885 lines (675 loc) · 24.7 KB

History

6.10.1 (2018-08-13)

  • Add per PMT contribution to tight coincidence level (#717)
  • Push git-tag before actual commit (#714)
  • clean map name and RunDB example (#715, #716)

6.10.0 (2018-07-19)

  • Clustering and desaturation upgrade for high-E (#712)
  • Enabling lone-hit simulation in FAX (#708)
  • notebooks to update RunDB and fix NN file version issue (#713)
  • Some minor improvement: Lower tight coincidence to 2-fold at pax level (#705), speed improvement for map interpolation (#676, #672), update S1 pulse shape parameter (#673, #665), show reconstructed position in WF plot (#677)
  • FIX related to Maps/inputs: 4th FDC map (#670), PMT after pulse (#671),maps renaming (#654), add missing tensorflow files (#679), new NN files after SR1 (#712)
  • Code style and dependencies: Issue template (#707), solfware dependencies (#701, #683), increase deletion timeout (#667)

6.9.0 (2018-01-18)

  • Clustering change: Tuning gap size parameters, using index_of_maximum instead of left_central (#642), configuration of tight coincidence (#644)
  • Maps changes: Alpha map using time depending 3d FDC for SR1 (#643), per-PMT S2 LCE map update (#646), True LCE map (#658), Update time-dependent 3D FDC maps for SR1 (#662), Add SR0 and update SR1 corrected LCE maps (#661), Update S2 xy map (#664)
  • Tensor Flow NN: 3D FDC for TFNN in SR0 (#656), Time-dependent 3D FDC for Tensor Flow Neural Net (#655), tensorflow based nn of SR0 (#651), tensorflow based nn (#650), New TF Json with bad PMT List (#659)
  • Bug FIX: Don't include inactive PMTs in S1 pattern fit (#663), Bugfix for AFT calculation (#657), * Enhencement: Refactor S1 AFT probability calculation (#660), Add SR-dependent configs from processing (#652), More reasonable electron lifetime default (#647), Add sr1 pmt after pulse configs from beginning/middle/end of the runs (#645)

6.8.3 (2017-09-26)

  • saving per channel contribution to tight coincidence (#641)

6.8.2 (2017-09-26)

  • New SR0 Net for SR0 reprocessing (#612)
  • Tuning clustering for SR1 (#615)
  • New maps for SR1 and SR0 (#625, #626 #628, #631)
  • Time dependent 3-D FDC (#632)
  • FAX update: single electron shape (#616), New PMT after pulse model for SR1 (#635), Update single-PE and S1 waveform model (#639)
  • No FDC in pax (#620)
  • Delete old maps (#627, #621, #623, #624)
  • Bug fixes (#618, #638, #633)
  • Code cleaning and style changes (#614, #619, #630, #634)

6.8.1 (2017-09-26)

  • New 3D data-driven FDC map (#611)
  • Fix de-saturation bug for very rare event (#601)
  • SR1 PMT gains set to default in fax config (#606)
  • S1 optical photon hit patterns with improved binning (#603)
  • New S1 area fraction top map (#598)
  • Prepare for supernova trigger (#568)
  • Don't keep LED data in buffer (#600)
  • Add random initial pickle to acquisition monitor file (#608)

6.8.0 (2017-08-07)

  • Two step gap clustering (#596)
  • Add data-driven SPE distributions for each channel (fax) (#595)
  • InterpolatingMap bug fix (#582)

6.6.6 (2017-06-28)

  • add S2 correction in task list (#573)
  • S1 area fraction top plugin (#572)
  • Update package versions in environment creation (#578)
  • New per-PMT LCE maps for full LXe+GXe volume (#577, #588)
  • Fax update for better photo-ionisation model and memory leak fix (#574, #581, #589)
  • Classification update (#576, #584)
  • Disabling pulse-based desaturation for S1 (#580)
  • New Position reconstruction (#591, #592)

6.6.5 (2017-05-04)

  • S2 map update for SR1 with new gains (#570)
  • S1 LCE map update for SR1 with new gains (#571)
  • Re-enable reduction of processed data (#569)
  • Coherent noise and maximum S2 afterpulse delay (#562)

6.6.4 (2017-04-24)

  • New Neural network training results with PMT 2 off in SR1 (#560)
  • Update XENON1T_MV.ini file for gain calibration (#561)
  • Field Distortion Correction and LCE maps for SR1 (#563)
  • New S2 correction map for SR1 (#565)
  • Add threshold on saturation correction reference region to avoid biased correction for S1 (#566)
  • Fix event builder error related to UTC timestamp (#567)

6.6.3 (2017-03-30)

  • Do not compute S1 AFT for large S1 (#558)
  • Don't delete area<0 hits in fixed range hitfinder mode (#557)

6.6.2 (2017-03-15)

  • Fix peak.hits_fraction_top calculation (#555)
    • Bug found in S1AreaFractionTopProbability calculation (#553)

6.6.1 (2017-03-11)

  • Updated field correction and S1 LCE map (#551)
  • Fix typo in S1 Area fraction top computation (#550)

6.6.0 (2017-03-08)

  • Updates to data files:
    • S2(x,y) relative light yield map (#535)
    • Neural net (#531)
  • S1 area fraction top probability computed (#533)
  • Fix bug in peak.area_midpoint computation
  • Simulator: Use realistic SPE spectrum in fax (#528)
  • Simulator: Fix bug in self-trigger emulation, AmBe noise file (#537)

6.5.1 (2017-02-28)

  • Minor bugfixes (#526, #527, #529)

6.5.0 (2017-02-21)

  • Change S1/S2 pairing loop order (#522)
  • Minor speed improvements (#520)
  • Add S1 hit-based pattern fit variable (#525)
  • Simulator: extend afterpules to just after the initial signal (#521)
  • Simulator: effective S1 shape model (#524)
  • Travis builds fixed (#523)

6.4.2 (2017-02-19)

  • Update author list
  • Fix drift velocity fetched in the runs database (no change in pax)

6.4.1 (2017-02-18)

  • Fix classification minus sign error
  • Fix waveform simulator internal bug

6.4.0 (2017-02-17)

  • New classification based on new properties. Order S1s by coincidence. (#510)
  • Cluster based on central regions of hits (#504)
  • Simulator: aft truth info, update for S2, afterpulse gains, fix dead PMT double counting (#516)
  • Simulator: Update SE gain, double PE emission (#507)
  • Update (r,z) correction map (#512)
  • Update S2(x,y) correction map, top and bottom correction map (#515)
  • Fix saturated PMT accounting config typo (#513)

6.3.3 (2017-02-14)

  • Fix for rare edge cases (#499)
  • ROOT-based event display (#503)
  • Fax update: afterpulses, gains, S2 AFT (#502)
  • Add run number to plot_dir filename (#500)

6.3.2 (2017-02-13)

  • Fix (r,z) correction bug (#498)
  • Neural network update (#495)

6.3.1 (2017-02-09)

  • Fix crash due to edge case due to negative samples in hits (#490)
  • Remove AWS support (#489)

6.3.0 (2017-02-08)

  • Hitfinder upgrade to accomodate long-tailed pe-pulses (#479)
  • r-z position correction from field simulation (#482)
  • Pulse-based desaturation correction (#480, #484)
  • New S1(x,y,z) correction map (#488) and QE rescaling fix
  • S2(x,y) correction (#487)
  • Realistic photoelectron pulse model (#478)
  • Peak property cleanup (#485)
  • Pickle output for plotting (#481)
  • Plugin to apply software prescale retroactively (#486)

6.2.1 (2017-01-11)

  • Fax: first realistic noise and ZLE thresholds (#471)
  • Fax: first realistic PMT after-pulse configuration (#475)

6.2.0 (2016-12-20)

  • Low-energy classification update, new peak properties (#467)
  • Corrected error in peak.mean_area_over_noise computation
  • Fix timezone offset for event time in plots (#466)
  • Pattern goodness of fit is now a likelihood ratio (#465)
  • Event builder robustness to db connectivity issues (#461)
  • Fixed-range hit finder (#458)

6.1.1 (2016-11-24)

  • Fax update - PMT and S2 afterpulses (#452)
  • Include factorial in posrec likelihood (#459)

6.1.0 (2016-11-11)

  • Improved clustering (#450)
  • Make processed data smaller (#456)
  • Write fax output to its own ROOT file (#453)
  • Neural network update (#457)
  • Fix for acquisition monitor pulse rescue logic (#454)
  • Bug related to decimal types in datastructure (#447)
  • Assorted event builder improvements after major changes

6.0.2 (2016-10-31)

  • More lenient timeouts for batch queue processing

6.0.1 (2016-10-31)

  • Fixed memory leak

6.0.0 (2016-10-26)

  • Remote/distributed multiprocessing (#439)
  • New S1 overall LCE maps (#436) and S2 per-pmt LCE maps (#431)
  • Amazon Dynamo support (#438)
  • Saturation correction computed, not added to total for XENON1T (#437)
  • Matched waveform simulator S2 model (#428)
  • Event display improvements (#440, #427)
  • Event builder speed improvements (#434)
  • Fax naming changes (#443)

5.7.0 (2016-10-07)

  • Eventbuilder improvements
  • Trigger base logic improvements (#407, 418)
  • do_not_trigger option (#419)
  • Fix for online dead time calculation (#416)
  • Trigger monitor improvements (#412, #424, #429)
  • Save acquisition monitor pulses (#414)
  • pulses_per_channel and n_pulses attributes added to Event class (#425, #422)
  • Fax saves GEANT4 id to ROOT file, more noise options (#426)

5.6.5 (2016-09-21)

  • Screwed up release, fixed HISTORY.rst

5.6.4 (2016-09-21)

  • Mistaken release

5.6.3 (2016-09-21)

  • MV processing support further

5.6.2 (2016-09-01)

  • MV processing support

5.6.1 (2016-09-01)

  • Muon veto processing support using run database information

5.6.0 (2016-08-31)

  • Deal correctly with gains almost zero but see hit (#415)
  • LCE map from Kr83m (#413)
  • Raw data from MV had suffix (#411)

5.5.1 (2016-08-15)

  • Fixes for trigger (#409)
  • Adjust drift velocity (#410)

5.5.0 (2016-08-08)

  • Muon veto integration (#406)

5.4.0 (2016-08-06)

  • New raw data format
  • Reconstruction fix (#401)
  • Event builder changes (#404)

5.3.6 (2016-07-25)

  • Small DAQ fixes.

5.3.5 (2016-07-25)

  • Trying to get a DOI

5.3.4 (2016-07-25)

  • Small tweak to description in setup.py (which I actually made because I want to get a DOI for pax).

5.3.3 (2016-07-25)

  • Small DAQ fixes.

5.3.2 (2016-07-20)

  • Only TPC errors can stop DAQ.

5.3.1 (2016-07-19)

  • Fix MV bug in ini (#397).

5.3.0 (2016-07-19)

  • Trigger changes for deadtime and acquisition monitor - no backword compatbility for new data (#395)
  • Muon veto DAQ changes (#396)

5.2.1 (2016-07-06)

  • Fix release

5.2.0 (2016-07-06)

  • Store acquisition monitor info (#392)

5.1.0 (2016-06-27)

  • S1 relative light yield map update (#382)
  • Makefile now writes to stable branch (#379)
  • Geant4 interface simulates from below cathode if desired (#389)
  • Trigger changes with error handling (#386)
  • Trigger prevent invalid event ranges (#388)
  • Equalized gains (#387)

5.0.2 (2016-06-20)

  • Unintentional release.

5.0.1 (2016-06-20)

  • Minor DAQ fixes including error handling (#384)

5.0.0 (2016-06-08)

  • Ready for XENON1T data
  • New clustering (#372)
  • Extended trigger window (#369)

4.11.0 (2016-06-06)

  • Run database interface (#366)
  • Revive PMTs that were masked (#371)

4.10.2 (2016-06-01)

  • ROOT fix (#370)
  • Configurable low-level info (#368)
  • Event builder fixes (scattered commits)

4.10.1 (2016-05-30)

  • Add PMT information (#364)
  • Event builder changes (#365 plus other commits), including processing related changes.

4.10.0 (2016-05-20)

  • Initial pax tuning for XENON1T #361

4.9.3 (2016-05-12)

  • Temporarily downgrade scipy due to issues with latest build on some systems
  • Event builder: split collections handling, save-all-pulses / mega event / timed trigger mode
  • Lowered threshold in XENON1T-LED config until we can specify optimal threshold (#357)
  • Waveform simulator bugfix (#354), LED signal simulation (#355)

4.9.2 (2016-05-03)

  • Poisson likelihood statistic for position reconstruction, confidence contour improvement (#342)
  • Event builder: parallel queries, delete-as-we-go, optimized queries, better config / run_doc handling
  • Lock-based race condition prevention for ROOT class compilation (see #351)
  • Fix wrong numbers in connector map (#349)

4.9.1 (2016-04-25)

  • Neural net uses correct QEs
  • Small changes for event builder
  • Split S2 afterpulse models so independent for XENON100 and XENON1T

4.9.0 (2016-04-18)

  • XENON1T: gains to 1 in LED mode, amplifiers and positions in pmts config dictionary (#339)
  • XENON100 S2(x,y) map, XENON100 S2 simulation bugfix (#334)
  • Event builder fixes, cax integration
  • Pax version no longer append to output filename (0f26ac0)
  • Multiprocessing and ROOT fix (#337)
  • Waveform simulator afterpulses fix (#341)

4.8.0 (2016-03-29)

  • New event builder version (#336)

4.7.0 (2016-03-21)

  • Geant4 input to waveform simulator
  • Tuning classification for XENON1T gas-mode zero-field.

4.6.1 (2016-03-07)

  • Screwed up release, fixing...

4.6.0 (2016-03-07)

  • Confidence levels on position reconstruction
  • Saturation correction bug
  • Several small bug fixes
  • Minor event builder changes

4.5.0 (2016-02-26)

  • .cpp classes now included within the ROOT output file (#323)
  • Area corrections stored separately in datastructure (#322)
  • Waveform simulator refactor, PMT afterpulses support (#321)
  • Small event builder changes (#316, several loose commits)

4.4.1 (2016-02-05)

  • Weird outlier bug fixes found in bulk processing

4.4.0 (2016-02-02)

  • New event builder iteration (#297)
  • Configuration bugs fixed

4.3.2 (2016-01-31)

  • Small argument fixes for default configuration.

4.3.1 (2016-01-28)

  • Nasty multiprocessing bug fix

4.3.0 (2016-01-25)

  • Parallelization refactor (#298)
  • Store meta data in ROOT output (#303)
  • z coordinate system now negative in liquid (#302)
  • Neural net reconstruction (#296)

4.2.0 (2016-01-11)

  • Fixes for ROOT output: memory leak (#282), LED output (#283), long int fields (#289)
  • Event builder changes (#278)
  • 3D position reconstruction for S1s (#277)
  • Hits and Pulses for S1s saved by default (#283)
  • Raw data filename format changed, progress bar fix (#289)

4.1.2 (2015-11-30)

  • Docs fixes
  • TableWriter bug
  • Saturation bug #274

4.1.0 (2015-11-17)

  • ROOT class output
  • Signal processing speedup (#245)
  • S1 3d pattern simulation & goodness of fit computation (#237)
  • Modifications for working with other TPCs (#247)
  • Improvements to / fixes for noisy channel hit rejection
  • Assorted bug fixes (#241, #244) and documentation fixes

4.0.1 (2015-10-17)

  • Memory leak fixed
  • Corrections to position reconstruction (#244)
  • Documentation fixes

4.0.0 (2015-10-02)

  • Add/remove several peak properties (#223, #214, #203), such as the peak's hits-only sum waveform.
  • Clustering changes: separate plugins, better goodness of split, faster (#223, #213)
  • Python 2 support (#217)
  • Paxer options to switch input and output type (#212)
  • Position reconstruction before classification (#223)
  • Fast PatternFitter for position reconstruction (#233)
  • Irregular correction map support, XENON100 S1(x,y,z) correction (#219)
  • S1 vs S2 classification fix (#221)
  • Several bugfixes and documentation improvements (e.g. #230)

3.3.0 (2015-08-03)

  • Natural break declustering (#187)
  • Improvements to chi2gamma accuracy and speed (#193, #196)
  • Non-continuous events in ZippedBSON format (#192)
  • XED writing (#177)
  • Refactor plugin base and timing code (#190)
  • S2 LCE in waveform simulator (#185)
  • Cleanup plugin folders and names (#202)
  • Minor improvements to logging (#155, #86) and plotting (#98, #144, #200)
  • Documentation improvements

3.2.0 (2015-07-06)

  • Multithreading of paxer (see --help)
  • Clustering bug fixed (#186)
  • Contribution section for non-XENON TPCs in examples.
  • Chi2 algorithm now runs by default (and has energy cutoff for speed)
  • Event builder pretrigger merged into pax
  • Units now statically defined
  • Various docs improvements

3.1.2 (2015-06-07)

  • Update requirements.txt
    • Require new numba version since use new features
    • Pymongo3 required for all our Mongo setups

3.1.1 (2015-06-07)

  • Fixed merge issue with minor release (mea culpa)

3.1.0 (2015-06-07)

  • Simplified hit finder (#167)
  • ZLE in waveform simulator
  • BSON output
  • Cleanup of Travis building
  • Various bug fixes

3.0.0 (2015-04-20)

  • Reprocessing capability, switch HDF5 backend (#116)
  • Better clustering algorithms: MeanShift, GapSize (#124)
  • Hitfinder: faster, new noise definition, work on raw ADC data (#126)
  • Bad channel rejection -> suspicious channel testing (#126)
  • ROOT output, including tests for Travis (#127)
  • Speed and feature improvements to folder-based IO (XED, Avro, ...) (#131)
  • Datastructure update (#139)
  • Bugfixes, plotting and comment improvements

2.1.0 (2015-02-15)

  • Avro raw data output
  • Neural net reconstruction
  • And lots of meaningless commits to get Travis continuous integration and Coveralls code coverage to work! (And ROOT, which will be in 2.2)

2.0.1 (2015-01-26)

  • Travis CI continuous integration is enabled
  • Minor bug fixes
    • Values missing from output if they were always default
    • Memory leak after many events due to logger

2.0.0 (2015-01-15)

  • Changes to core
    • Cleanup of datastructure (see #80 and #81)
    • Most of core wrapped in Processor class
    • Plugins shut down at end of Processor.run(), not just on destruction
  • New signal processing chain
    • BaselineExcursionMethod, finds single-photon peaks in each channel (Default)
    • FindBigPeaks, a traditional sum-waveform peakfinder
    • Supporting peak classification and property computation plugins
  • Chi-square gamma x,y position reconstruction
  • Waveform simulator enhancements
    • Wrapped in Simulator class, loaded along with processor
    • Performance improvements
    • Basic zero-length encoding emulation
  • WritePandas: write our data as DataFrames to containers supported by pandas
  • 2D channel waveforms plot
  • Support for arbitrary external detectors / extra channel groups
  • More tests

1.4.0 (2014-11-21)

  • DSP
    • Peak width fields added to datastructure
    • newDSP: Interpolated peak width computations
    • DSP plugins cleaned up and reorganized (except old peak finder)
    • Frequency bandpass filtering support
  • Updated docs, comments, logging
  • Plots
    • 3D channel waveforms plot
    • Event summary plot
  • Music output (fun side project)
  • Separate directory for example data files
  • Configurations for XAMS, Bern test setup
  • --input and --output override settings for most plugins
  • WaveformSimulator: improved defaults
  • Stable DAQ injector
  • Various bug fixes and cleanups that polish

1.3.0 (2014-10-17)

  • Plugin directory moved (fix bug in previous release)
  • Bulk processing enhancements
    • Scripts for parallelization
    • XED: read in entire datasets, not just single files
  • More command line arguments: input, plotting
  • Configuration enhancements
    • Module-level settings
    • Multiple inheritance
  • DAQInjector
    • New run-database format
    • Repeat single events
    • Create shard index
    • Further debugging and maturing
  • Implement run database interface
  • WaveformSimulator (Fax) cleanup:
    • Several truth file & instruction file formats
    • Better-motivated settings
    • ER/NR S1s
  • Cut overhanging pulses
  • Several PosSimple improvements
  • Interpolating detector maps (for position-dependent signal corrections)
  • Plot 2D hit patterns

1.2.0 (2014-10-02)

  • DAQ injector - can inject data into DAQs.
  • Nested configurations - better handling of configurations and allows for nesting

1.1.0 (2014-08-29)

  • HDF5 output - will be, for now, default output format.
    • We now have a binary output format for peaks and event.
    • Should also be easily extendible to ROOT output, which is blocked until ROOT solves some Py3.4 bugs.
    • Allows bulk comparison with high statistics for things like trigger efficiency
  • Bug fixes relating to difference between XENON100 and XENON1T formats (occurences extending past event windows).
  • Starting work on a new SimpleDSP processor
  • Waveform generator
  • General bug fixes and cleanup

1.0.0 (2014-08-16)

  • Completely refactored event datastructure
  • Moved from Python dictionaries to an event class, seen in pax.datastructure
  • Ported all modules with pax to the new structure
  • Should open up I/O and C++ binding opportunities
  • Now there are Event, Peak, Waveform, and ReconstructedPosition classes
  • All of this is based on an extensively modified fork of 'micromodels'.
  • Input control (See Issue #26)
  • Can now run pax with single events
  • Run paxit --help to see how one can process events
  • Binaries of paxit installed when pax is installed
  • Improved testing
  • Started testing plugins (this will start including other plugins later in the release)
  • Extensively testing the event class
  • Peak finder now nearly identical to Xerawdp: better than 99.9% agreement on >20000 peaks tested
  • Simulation of the Xerawdp convolution bug (filtered waveform mutilation around pulse edges)
  • Small bugfixes (empty isolation test regions, strange behaviour when max of filtered waveform is negative)
  • Xerawdp XML file interpretation is off-by one (min_width=10 means: width must be 11 or higher)
  • Integration of a waveform simulator (FaX) which can simulate S1s, S2s, and white noise
  • Script to convert from MC/NEST root files to FaX instructions
  • Simplified but much faster simulation mode used for peaks >1000 pe
  • Plotting improvement: largest S1 & S2 in separate subplot
  • Numerous bug fixes:
  • Pickler I/O
  • Remove dead code (clustering)

0.2.1 (2014-08-14)

  • paxit binaries installed by default to allow working out of source

0.2.0 (2014-08-04)

  • Define static event class data structure
  • Transforms now specified in ini file
  • Can launch small web server for viewing plots
  • Major changes to the peak finding to better match Xerawdp. Agreement is currently at the 95% level.
  • Two important bugfixes for determining included channels : XED channel mask parsing, 0->1 start
  • Filter impulse response now identical to Xerawdp
  • Different summed waveforms for s1 and s2 peakfinding
  • Transforms have start and stop methods

0.1.0 (2014-07-18)

  • First release of software framework
  • Functional but not complete digital signal processing * Sum waveform for top, bottom, veto * Filtering with raised cosine filter * Peak finding of S1 and S2
  • Basic inputs * MongoDB (used online for DAQ) * XED (XENON100 format)
  • Basic outputs * ROOT * Pickle * Plots
  • Demo reconstruction algorithm of charge-weighted sum