Release Notes¶

main¶

Oct. 24, 2024 don la dieu (nega at icecube.umd.edu)¶

IceTray Release v1.12.0

make virtual function non-virtual, fixing a downstream class (#3533)

June 18, 2024 don la dieu (nega at icecube.umd.edu)¶

IceTray Release v1.11.0

Ensure that extension modules import cleanly (#3493)
Run mypy in CI (#3480)
Add stubs for even more pybindings (#3413)

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)

Dec. 14, 2023 don la dieu (nega at icecube.umd.edu)¶

IceTray Release v1.9.0

Remove Uber Header (I3.h) (#3151)
python ruff cleanup fix EXE001 (#3289)

Sep. 28, 2023 don la dieu (nega at icecube.umd.edu)¶

IceTray Release v1.8.0

Fix typo in an error messages

June 22, 2023 don la dieu (nega at icecube.umd.edu)¶

IceTray Release v1.7.0

Make I3Tray.I3Tray and icecube.icetray.I3Tray the same thing. (#3102)
Remove all references to load_pybindings (#3086)
General code cleanups
Improve tests’ parallelism

Apr. 12, 2023 don la dieu (nega AT icecube.umd.edu)¶

IceTray Release v1.6.0

Cleanup temporary file use in tests
Minor cleanups

Dec. 12, 2022 don la dieu (nega AT icecube.umd.edu)¶

IceTray Release v1.5.0

Reintegrate SegmentedSplineReco (#2630)
Documentation cleanups

Jul. 27, 2022 don la dieu (nega AT icecube.umd.edu)¶

IceTray Release v1.4.0

Drop support for Python 2
Clean up documentation

December 16, 2021 Alan Coleman (alanc@udel.edu)¶

IceTray release v1.3.0

Changed compiler compatibility checks for GLIB

Dec. 20, 2019 Alex Olivas (aolivas@umd.edu)¶

Combo Release V00-00-00

December 13, 2019 Alan Coleman (alanc@udel.edu)¶

Release V01-11-02

Extended pybindings to expose all functions for LogLikelihood and SeedService
Added python sub-directory so that gulliver can be loaded by itself

December 1, 2017 Kevin Meagher¶

Release V01-11-01

add Print() function to I3LogLikelihoodFitParams
docs cleanup

March 23, 2017 EKB (blaufuss@umd.edu)¶

Release V01-11-00

updates to use icetray::serialization over boost::serialization
docs cleanup

April 3, 2015 Meike de With (meike.de.with@desy.de)¶

Release V01-09-00

More python bindings for gulliver, so we can have gulliver modules entirely in python.
Clean out some old ROOT fossils (linkdef, rootcint).

– David Boersma

December 9, 2014 Meike de With (meike.de.with@desy.de)¶

Release V01-08-03

Adding a lot of documentation.

– David Boersma

September 29, 2014 Meike de With (meike.de.with@desy.de)¶

Release V01-08-02

Update documentation url.

– Don la Dieu
Start RST documentation.
Enable both context names and objects as input for LLH services.

– David Boersma

July 11, 2014 Meike de With (meike.de.with@desy.de)¶

Release V01-08-01

Bring pybindings up from subdirectories.

– Jakob van Santen
FastLogSum bugfix (use ieee754 without casting).

– David Boersma

April 8, 2014 Meike de With (meike.de.with@desy.de)¶

Release V01-08-00

Add pybindings for GetDiagnostics.
Add const-correctness to I3LogLikelihoodfit.
No longer check for one million radians, as AnglesInRange takes moduli.

– Jakob van Santen
Make implicit conversions to bool explicit.

– Chris Weaver
Add a test to deal correctly with directionless particles.
Reduce logging level.

– David Boersma

January 21, 2014 Meike de With (meike.de.with@desy.de)¶

Release V01-07-00

No longer require the icepick project.
Make it possible to implement I3MinimizerBase in Python.

– Jakob van Santen

June 29, 2013¶

Release V01-06-05

Add pybindings for I3EventLogLikelihood.
Pybindings give clearer information about unimplemented methods.

February 21, 2013¶

Release V01-06-04

Removed dependency on log4cplus.
FastLogSum and ieee754 have been moved from phy-services to gulliver.
Use list_indexing_suite instead of std_vector_indexing_suite.
Added tracing functionality: store all parameter values (and corresponding function values) that were tried by the minimizer during a fit. This trace (if enabled) is stored for every fit as an I3VectorDouble object.

September 11, 2012¶

Release V01-06-03

Use GetName() instead of getting a configuration from the context.
Allow gradients to pass through I3EventLogLikelihoodCombiner cleanly.
Update the API in I3TestDummyEventLogLikelihood to reflect changes to gradient handling in Gulliver.

June 25, 2012¶

Release V01-06-02

Add keyword args to I3EventLogLikelihood::GetLogLikelihood pybindings.
make gradients optional.
Remove reference to I3ConverterFactory.
Allow SeedServices to be implemented in Python.
Added pybindings for I3Vector<I3LogLikelihoodFitParams> and setting the geometry.
Make parametrizations implementable in Python.

January 11, 2012¶

Release V01-06-01

Expanded pybindings allow Python modules to use Gulliver services.
Provide SetGeometry method.
Root dependency is now optional.

August 09, 2011¶

Release V01-06-00

Get diagnostics from all 3 services.
Add new virtual method: SetEvent(frame) and GetDiagnostics(eh).
Add a new function double I3EventLogLikelihoodBase::GetLogLikelihood( const I3EventHypothesis &ehypo, I3EventHypothesis gradient, bool maximize_extra_dimensions) that can be overridden by fancy likelihoods. The default implementation calls through to the old GetLogLikelihood() or GetLogLikelihoodWithGradient(). Likelihoods that receive this call with maximize_extra_dimensions=false should _not_ modify the hypothesis, even though nothing prevents them from doing so. I3Gulliver calls this with maximize_extra_dimensions=true as appropriate for a fitter. I3LogLikelihoodCalculator retains the expected behavior.
Make tableio converters DEFAULT (October 29, 2010).
Reduce verbosity.
Add tableio converter (June 4, 2010).
Add PLogL filter.
Some documentation updates (October 6, 2009).
Moved dummy service implementations (for unit tests) from public/gulliver/test to the private/test directory.
Added support for gradients (October 6, 2009):
- The minimizer decides whether to use gradients or not. If you write a minimizer that uses gradients, it should call the gulliver functor with two vectors of doubles (one for the parameters, one for the gradient) instead of two. You also need to overload the ‘bool UsesGradient() const’ method, it should return true.
- The I3GulliverBase and I3Gulliver classes got one extra method, the operator() method with two vector refs (one const, one non-const for the gradient) instead of two. It checks that the likelihood & parametrization services indeed support gradients (if not: log_fatal) and connects the tubes and wires between them. Of course it multiples the gradient with -1, because it also does that with the llh function value.
- The I3EventLogLikelihoodBase class has now two new methods: double GetLogLikelihoodWithGradient(const I3EventHypothesis &h, I3EventHypothesis &g ) which gets called in case gradients are wanted. The g argument serves both to specify which derivatives need to be computed (all datamembers are initialized to zero except those for which derivatives need to be computed) and to store the values of those derivatives. bool HasGradient() should be implemented to return true if the service can indeed compute gradients.
- The parametrization class must perform a chain rule: it initializes the gradient vector (with 1s and 0s, to order the gradient in terms of physics variables) and from its values (computed by the likelihood service) it transforms that to the space of minimizer parameters. Two new methods in I3ParametrizationBase: bool InitChainRule(bool) informs the parametrization service whether gradients are going to be wanted (bool arg) and if that is the case, then the parametrization service should do the necessary initialization (e.g. initialize the gradient object), and return true (default implementation returns false to signify that gradients are not supported).
- TODO: a test factory, a utility that computes derivatives both numerically (using GetLogLikelihood()) and analytically (using GetLogLikelihoodWithGradient()) and checks that they are the same; directly from the likelihood service and indirectly through the parametrization service.
- NOTE: there are not real life implementations yet of this, e.g. in lilliput; I3SimpleParametrization still needs chain rule implementation, someone should make a wrapper for some gradient-capable minimizer (both GSL and ROOT/Minuit can do that, I think). Nathan (who triggered this whole gradient effort) is working on splinified photorec, so photorec should soon have derivatives.

August 11, 2009¶

Release V01-04-00

Python bindings

October 27, 2008¶

Release V01-03-00

Added LogL filter.
Added LogL ratio filter.
Release made for std-processing.

August 19, 2008¶

Release V01-02-01:

Minor updates, (CMakeLists.txt tweak, python stuff).
Release made for icerec purposes.

June 24, 2008¶

Release V01-02-00

Got python bindings of I3LoglikelihoodFitParams, for loot.
AnglesInRange now uses fmod instead of a while loop should waste less time on confusing near-vertical events with crazy azimuth result.
Added name/label in I3Gulliver, used in log messages.
Moved ‘minimum ball’ utility to phys-services.

January 3, 2008¶

Release V01-01-00

Added a ‘minimum ball’ utility, a dynamic geoselector which selects only the DOMs within a ‘minimum ball’ around the DOMs that have hits/pulses. Useful for phpnh-like likelihood functions, such as WFllh and photorecllh.
AnglesInRange(p) check: only if particle has a direction; and if it has, but it’s impossible to force it in the normal range, then declare the fit failed.
Release for the benefit of online filters.
Note: the “minimum ball” and direction-related utilities are temporary in this project. They are also available in the current trunk of phys-services. As soon as a new phys-services release is used within icerec, these utilities should be removed from gulliver/trunk and a new gulliver point-release should be made without them.

October 31, 2007¶

Release V01-00-04

Changed behavior in case the minimizer ends with pathological angles (zenith or azimuth larger than 1e6). then just decide to proclaim the fit ‘not converged’. this is good for pole filtering stability, among others. WIMP people need to be aware of this, because the pathological angles are more likely for near-vertical events.
Add utility functions with I3Direction objects.
const-ified GetName() method of I3ParametrizationBase.

September 18, 2007¶

Release V01-00-03

Fix a minor but annoying bug: sometimes (rarely) the minimizer makes the zenith/azimuth angles so large that 2pi is less than the floating point accuracy of this number. Happens about 2-3 times a week during online filtering, and then stuff hangs (infinite loop). Fixed now.

September 13, 2007¶

Release V01-00-00

Event loglikelihood combiner: default relative weight is 1.0 (was: 1/N, for N combined likelihoods).
Much more documentation.
Underscore suffixes for datamembers (code review).
Remove obsolete/unused SimpleStatistics class.
Add “ordinal” utility function, for log_messages (was a static function in several modules and services).
Add “unit test constructor” to I3ServiceBase (sets name directly, makes a dummy I3Context and deletes that context at destruction time).

August 10, 2007¶

Release V00-05-02

Some convenience methods here & there.
Implementation of part of the code review suggestions.

April 23, 2007¶

Release V00-05-01

Added likelihood combiner service (for Bayesian reconstruction).
Introduced templated factory classes for simple services. These eliminates the need for bulky factory classes which duplicate half of the code of a simple service. There are two versions: one for service which need a unique instance for every context, and one for services for which a single service object can be used without problems in many contexts.
Added utility directory. Some of that stuff might be useful for other projects as well, but then it it should move to a more central project.
The seed service interface now contains more virtual methods, through which a “user” (typically somebody coding an I3Module) can prepare a new seed from some modified first guess. Here we assume the following terminology:
- First guess: e.g. a linefit, JAMS, cfirst result or just a I3Particle/I3EventHypothesis object constructed by the module from some other result or just the data.
- Seed: the starting I3EventHypothesis object which has meaningful values in all datamember (most first guesses e.g. lack an energy estimate). For event types other than muons & cascades you probably need to define an appropriate I3FrameObj.
The I3LogLikelihoodFit now contains a I3EventHypothesis and a I3LogLikelihoodParams (was: a I3ParticlePtr, a I3FrameObjPtr and a I3I3LogLikelihoodParamsPtr).

Agust 29, 2006¶

Release V00-02-00

Renamed I4Particle to I3EventHypothesis.
More ground work on I3LogLikelihoodFit and I3Gulliver (inspired by requirements for iterative fit and paraboloid).