Release Notes¶
main¶
Feb. 20, 2024 don la dieu (nega at icecube.umd.edu)¶
IceTray Release v1.9.2
Delete mailinglist files (#3357)
Add upgrade rules to ruff (#3316)
Remove all tab characters used for indentation from python files (#3300)
Dec. 14, 2023 don la dieu (nega at icecube.umd.edu)¶
IceTray Release v1.9.0
Clean up invalid escape sequences for python3.12 (#3286)
Fix E731 lambda-assignment (#3271)
general python cleanups (ruff/E713) (#3269)
Fix a number of issues identified by ruff (#3262)
remove unnecessary __future__ imports (#3243)
Relax test tolerance on Apple Silicon (#3213)
general python cleanups
Oct. 12, 2023 don la dieu (nega at icecube.umd.edu)¶
IceTray Release v1.8.2
update millipede to use Kahan summation routine for logl calculation (#3196)
Sep. 28, 2023 don la dieu (nega at icecube.umd.edu)¶
IceTray Release v1.8.0
replace deprecated unittest asserts with modern equivalents (#3168)
June 22, 2023 don la dieu (nega at icecube.umd.edu)¶
IceTray Release v1.7.0
Remove all references to load_pybindings (#3086)
General code cleanups
Dec. 12, 2022 don la dieu (nega AT icecube.umd.edu)¶
IceTray Release v1.5.0
Reintegrate SegmentedSplineReco (#2630)
Jul. 27, 2022 don la dieu (nega AT icecube.umd.edu)¶
IceTray Release v1.4.0
Drop support for Python 2
Shared memory fixes for tests
Clean up tests
Refactor handling of configuration in I3ServiceBase to be clearer and less error*prone
Updates for I3Minuit / iminuit
Feb. 21, 2022 Tianlu Yuan¶
IceTray release v1.3.0
Remove DOMEfficiency parameter. Now handled via SetEfficiencies in photonics-service (#2809)
Relative uncertainty parameter in Millipede llh (#2788)
Remove bounds on L-BFGS-B in MonopodFit
Update tolerance in millipede test
Fix NameError bug (#2785)
Fix tests
Bound monopod zenith only if using L-BFGS-B
Increase default PCG iterations from 100 to 1000; add early exit if convergence is released (#2599)
Change PCG tolerance from 1e-8 to 1e-10
Dec. 20, 2019 Alex Olivas (aolivas@umd.edu)¶
Combo Release V00-00-00
(r171368): Enforce minimum bin width in BinSigma mode
(r171365): Fix an initialization bug in bayesian blocks rebinning
(r170477): Use all OMTypes except IceTop and scintillators (Upgrade support)
(r170476): Promote invalid DOM noise rate to a fatal error
June 12, 2018 Kevin Meagher¶
Release V02-01-06
make the LengthBounds parameter accessible in the taupede segment
November 29, 2017 Kevin Meagher¶
Release V02-01-05
- (r156726): Require user to explicitly set muon/cascade spacing to zero
if no Muon/Cascade PhotonicsService is provided (#1968)
Add Boundry parameter to millipede tray segment
documentation updates
python3 updates
script cleanups: remove Finsih and TrashCan
March 24, 2017 Mike Richman (mike.d.richman@gmail.com)¶
Release V02-01-04
(r151307): Update tests to handle 2016 GCD (#1896)
(r149760): Raise an error if a pulse starts _exactly_ at the end of the readout window (#1849)
(r149503): Support multi-particle hypotheses in artist
(r148349): Abort minimization for alpha<0 rather than crashing (#1773)
- (r146738): Fix bin counts when removing zero-width pulses (#1712)
– Jim Braun
May 2, 2016 Alex Olivas¶
Release V02-01-03
(r142136): Allow selection of photon tables from the GUI (#1249)
No longer have to import icetray and gulliver, because you need their base classes. – Jim Braun
April 3, 2015 Meike de With (meike.de.with@desy.de)¶
Release V02-01-02
Correct multiplicity calculation in MillipedeLikelihood account for empty bins. The reduced log-likelihoods should now be slightly less insane.
More helpful error message when Monopod is seeded with a track seed
Small code changes to fix warnings – Jakob van Santen
Recognize I3_TESTDATA
Fix tests – Alex Olivas
November 20, 2014 Meike de With (meike.de.with@desy.de)¶
Release V02-01-01
Fix to correctly handle events where a pulse has zero width, which can happen for sdst events that have two pulses extracted within 0.5 ns of eachother – Leif Raedel
September 23, 2014 Meike de With (meike.de.with@desy.de)¶
Release V02-01-00
Add Steamshovel artist
Fix memory-reuse bug in MillipedeDOMCacheMap – Jakob van Santen
July 11, 2014 Meike de With (meike.de.with@desy.de)¶
Release V02-00-01
Add some documentation of the photorec segment – Chris Weaver
April 22, 2014 Meike de With (meike.de.with@desy.de)¶
Release V02-00-00
Make new tests work with Python 3
Make conditional execution of Millipede segments work
Switch from NMML to PCG as a minimizer (now feature complete and better than NMML in every way!) and solve all related problems. High regularization sometimes slows convergence (it goes into the linear regime), but no pathologies seem to result
Improve numerical properties of the energy optimization in PCG
Fix segfault when gulliver hypothesis misses the detector and gradients are in use by just returning the noise likelihood rather than trying to create 0-width matrices
Add an experimental prior option for the Bayesian binning – Nathan Whitehorn
Make cholmod_sparse->I3Matrix conversion (and back) actually work
Query noise vector in PyPyMillipede
Pre-condition the gradient descent vector in the first iteration
Explicitly store the solution with the best likelihood, not just the smallest gradient (not necessarily equivalent)
Correct sign error which effectively set MinWidth to 200, which was surely not the intention
Store vector of particles for fit Foo as FooParticles instead of FooParams (which is not descriptive)
Use segments in the examples
Fill and return a real MillipedeFitParams instead of a fake one
Fix a variety of bugs in the handling of partial exclusion windows (promote exclusion-window insertion to a function, increase the size of raw_bin_edges and raw_charges, prevent invalid bins from absorbing valid ones in rebin_min_entries(), only consider exclusion windows that overlap with the readout window) and add clauses to the binning unit test to exercise the above
Add shims for named tuples and math.isnan() for Python <= 2.6 – Jakob van Santen
January 21, 2014 Meike de With (meike.de.with@desy.de)¶
Release V01-09-00
Teach PyPyMillipede how to evaluate gradients
Re-enable and extend pulse binning test
Handle an edge case that caused duplicate bin edges
Refactor pulse binning code such that user-configurable rebinning can be treated separately from exclusion window rebinning
Add tests for the energy solver and MonopodFit
Add an optional Bayesian Blocks binning mode (can be useful in situations where speed and reliability are more important than maximum best-case precision)
Add option to use LBFGS in MonopodFit
Use gulliver to put a custom FitParams in the frame instead of running another Monopod module
Integrate Patrick’s Taupede parametrization and fit
Add decorator to turn a segment containing a parametrization into a full-blown Millipede fit segment
Add option to use multiple seeds
Make binning in tests slightly coarser, for more convergence – Jakob van Santen
Make the default parametrization and iteration count consistent: use the simple parametrization by default – Nathan Whitehorn
August 21, 2013 Meike de With (meike.de.with@desy.de)¶
Release V01-08-00
Improve pybindings for PyPyMillipede – Jakob van Santen
Update angular reconstruction examples to work with new code
Fix errors in azimuthal gradient calculation in MuMillipede – Patrick Hallen
Make project compatible with Python 3
Use i3_assert() for data quality checks so that they are also run in release builds – Nathan Whitehorn
Add more documentation for monopod tray segment – Laura Gladstone
June 29, 2013 Emanuel Jacobi (emanuel.jacobi@desy.de)¶
Release V01-07-00
Make the exclusion strategy used for the HESE analysis canonical.
Add PyPyMillipede, a MillipedeService that can be instantiated and poked at from Python, for instance to power a steamshovel artist.
When re-binning pulses to satisfy the PEPerBin threshold, stop merging bins if the duration of the new bin would exceed 200 ns.
This fixes a regression introduced in r91790 that caused the re-binning loop to condense the time bins for DOMs with total charges less than PEPerBin into 2 bins: one empty bin extending from the beginning of the readout window to the start of the first pulse, and one containing the total charge and extending from the start of the first pulse to the end of the readout window. This bug severely degraded the performance of a single-source vertex fit on events with < 10 TeV deposited energy, but was much less noticable for high-energy events with many DOMs above the PEPerBin threshold, and thus went unnoticed for 9 months.
Add pybindings for MillipedeDOMCache.
Fixed some smaller bugs and compiler warnings. – Jakob van Santen
Fixed segfault for free cascade initial direction.
Fixed gradients. – Moriah Tobin
Code cleanups, fixed tests – Nathan Whitehorn
May 8, 2013 Emanuel Jacobi (emanuel.jacobi@desy.de)¶
Release V01-06-01
Fix a subtle bug for tracks that do not actually intersect the tracking volume where, instead of returning immediately, the code would try to solve for a very very very long track.
Better input validation
Tests use I3MCPE instead of I3MCHit.
Code cleanup – Nathan Whitehorn
Step sizes for monopod gulliver fit are accesible as parameters from tray segments. – Marcel Usner
February 21, 2013 Emanuel Jacobi (emanuel.jacobi@desy.de)¶
Release V01-06-00
Allow initial cascade direction to vary for starting or contained tracks in MuMillipedeParametrization
Treat starting/stopping/contained/infinite tracks with their respective shapes in MuMillipede
Update default DOM efficiency to 0.99
Use SPE scale from dataclasses
NOTE: If you have been manually setting the DOMEfficiency to include the 0.85 PE SPE scale, you MUST remove it as it is now always included (and may vary DOM-by-DOM in the future if we get around to measuring it)!
NOTE: This also means that all-default settings should produce correct energies with no required scalings or parameter changes. If you find you require any scalings, this is a bug – please report it.
Fix bug where NMML may (rarely) fail to converge and return a nonsense answer. Note that only the nonsense has been fixed; the circumstances under which NNLS is a terrible Poisson-likelihood seed have not yet been identified
Add tray segments that emulate the behavior of some older reconstructions (e.g. Photorec) using Millipede
The base class of MuMillipedeParametrization has been changed to I3SimpleParametrization, which may require some changes to option names in scripts using MuMillipede as a Gulliver service.
December 3, 2012 Nathan Whitehorn (nwhitehorn@icecube.wisc.edu)¶
Release V01-05-00
Fix Fisher matrix calculation to actually work
Replace NNLS + BFGS2 tuneup with NNLS + NMML for heavy lifting, providing much more accurate answers
Fix discrepancy between number of degrees of freedom calculated internally and reported to Gulliver
Improve error checking, quieting some superfluous warning messages
Require time window presence in frame for estimating readout window length
Add regularization support back after updates to solve maximum likelihood explicitly
Fix infinite loop possible in certain rare circumstances: convergence is now fully deterministic
September 26, 2012 Emanuel Jacobi (emanuel.jacobi@desy.de)¶
Release V01-04-01
Make Monopod instantly return for failed seeds.
Support new I3FrameObject: I3Matrix they constructible from numpy arrays via the array protocol.
Fully implement Fisher-matrix calculation for energy solutions in a new stand-alone module – Jakob van Santen
Fix for calculating time ranges from event headers
Instead of taking the calibration errata, saturation errata, an bad DOMs list as separate argument, take a single vector of lists-of-things-to-exclude.
Add option to use time window ranges as meant or to totally exclude DOMs appearing in the lists.
Define 1-cableShadow as DOM efficiency – Nathan Whitehorn
September 11, 2012 Emanuel Jacobi (emanuel.jacobi@desy.de)¶
Release V01-04-00
Merged with fortinbras branch new features include:
Support for masking out arbitrary time windows, e.g. clipped bits of the FADC.
Gradients will now pass through I3EventLogLikelihoodCombiner.
MillipedeFitParams::logl_ratio, the logarithm of the ratio of the best-fit likelihood to the maximum possible likelihood given the data. Unlike rlogl, this should be chi2-distributed even with small counts.
Pretty-printing for MillipedeFitParams.
Get the range of possible pulse times explicitly rather than relying directly on I3EventHeader. The fallbacks to the trigger window and range of pulse times are now hidden inside of MillipedeBase. Note that because I3TimeWindow is not actually a frame object in the current dataclasses release, the fallback will always be taken for now. – Jakob van Santen
Allow importing millipede to work even if the debugger’s dependencies aren’t satisfied.
Fix various mismerges and style errors
Provide a fallback for adding infinite exclusion windows for event with old-style calibration errata
Rewrite UpdateData() from scratch to properly handle exclusion windows.
Initialize logl_ratio properly.
Restore ability to skip unhit DOMs.
Totally ignore DOMs with no valid time windows (Bug reported by Tom Feusels)
For tracks that did not intersect the volume (i.e. tracks with no segments) do not try to treat the root track as a segment but fail gracefully.
Various bugfixes – Nathan Whitehorn
August 04, 2012, Emanuel Jacobi (emanuel.jacobi@desy.de)¶
Release V01-03-00
Support for LLH Gradients
Better protection against negative energies in the calculation of energy losses
Add sanity checks on the input pulse stream
Smoother handling of the case of zero sources
Test the maximum likelihoods in the pymillipede test to be maximal
Remove SingleEnergyGradient
Deactivate muon and shower regularization – Nathan Whitehorn
Allow parameterization of starting tracks in MuMillipede – Claudio Kopper
Optionally use a slant depth binning – Tom Feusels
June 25, 2012, Emanuel Jacobi (emanuel.jacobi@desy.de)¶
Release V01-02-00
Fix PhotonsPerBin = 0, which had ended up turning on amplitudes-only mode at some point and would otherwise have relied on memory corruption to function.
Work around broken GCD files in IC86 processing
Remove dependency on photonics
Refactor code to reduce code duplication. Note that this changes the options and default behavior of Monopod, which now uses PhotonsPerBin=5 as a default.
Update example scripts to new photonics-service API
Add Multidimensional maximum likelihood.
Compute the negative log likelihood from the beginning. – Nathan Whitehorn
Add a debugging callback to PyMillipede
Re-jigger pulse-binning loop so that bins no longer absorb long runs of zero charge. A zero-charge bin will now be added if the gap between two pulses is more than 10 bin widths.
Remove reference to I3ConverterFactory – Jakob van Santen
Make using unhit DOMs optional, default behaviour should stay unchanged. – Tom Feusels
March 6, 2012, Emanuel Jacobi (emanuel.jacobi@desy.de)¶
Release V01-01-00
Make the tolerance in SolveNoisyPoisson settable, and set it to a sanely small value. This signficantly improves performance on low-energy, noise-dominated events where the first-order solution is off by an order of magnitude or more.
Make
cableShadow_settable in Monopod – Jakob van SantenFix unit tests – Nathan Whitehorn
February 7, 2012, Emanuel Jacobi (emanuel.jacobi@desy.de)¶
Release V01-00-00 - Initial Release
copied from sandbox/nwhitehorn/millipede
