Releases: gundam-organization/gundam
Releases · gundam-organization/gundam
Gundam v1.4.3
What's new?
- Attempt to fix random segfault occuring on CCLyon
- Adding
selectionCutFormula
option in dataset definition - Now showing CPU efficiency while fitting
- Adding the option "-a" to force loading asimov in data container (regardless of
selectedDataEntry
andselectedToyEntry
) - Fixing memory leaks spoted by valgrind
Gundam v1.4.2
Gundam v1.4.1
What's new?
- Increase breakdown stat output
- gundamFitter now handles
-s
for specifying a given seed while using gRandom - Now xsllhFitter apps and libs aren't built along with GUNDAM sources. One can trigger back the building using cmake option
-D WITH_XSLLHFITTER=ON
- Added proper option
-D USE_STATIC_LINKS=ON
to enable static link - Added doxygen build setup: add
-D BUILD_DOC=ON
to cmake and thenmake doxygen
- Dropping version number in built libs
Bug fixes
- Fixed mirror dials
- Fixed bug while locking threads in SplineDial evaluation
- Fixed possible bug while indexing binned dial
Optimizations
- Optimized CPU parameter propagation (using embedded for_each)
- Using memory proximity data (RAM access is not parallelized)
- Lighter event loading
- Using DialWrapper to properly handle unique_ptr
- Various fixes/improvements in the toolbox
Gundam v1.4.0
What's new?
- Dataset definition now extended (proper support for Data, Fake-data, Toy-MC) with override options for custom variable definition
- Throwing statistical uncertainties on pseudo-data when using --toy option
- Adding support for custom toys definition using
--toy N
, where is the toy index. This is providing a selected index that is used to select a given variable array index while reading the input TTrees. - Lower RAM needed to load dials
- Adding CUDA support for GPU extension!!
- Adding dockerfile
- Adding
gundamFitCompare
which compares 2 output files fromgundamFitter
Full changelog of 1.3.5-> 1.4.0
Config fileset for T2K OA2020-like (not including inputs)
GUNDAM is comming to the Nintendo Switch!! 😉 (demonstrating the GPU capabilities!)
Gundam v1.3.5
What's new?
- Fixing YAML cpp lib link when using the custom made .cmake find package script
- The option
-D COLOR_OUTPUT=OFF
can be set in the cmake config to disable any colored output. - Using proper mutex for the spline eval
- Generalizing
minimumSplineResponse
tominDialResponse
andmaxDialResponse
- Extending the definition of
applyCondition
to a list definition underapplyConditions
which can be set with different syntax - Fixing error evaluation in sample bins
- Introducing custom LLH
BarlowLLH_OA2020_Bad
which reproduces the bug from the OA2020 in BANFF
Gundam v1.3.4
What's new?
- Adding
gundamConfigCompare
which scans 2 config files and highlight the differences - Added the
cmake/gundam-setup.sh
env script which provide an easy interface for compilation - Overall reshape of the CMakeList config files
- Showing IO read speed while loading the event to RAM
- Dial is now a pure virtual class
- Generalizing
minSplineResponse
tominDialResponse
andmaxDialResponse
- Rescaling plot frames wrt legend size
- Prefit/postfit comparison plot also generated on a normalized scale (free parameters are rescaled wrt the provided step/stdDev if available)
Full changelog of 1.3.3-> 1.3.4
Preview of the new app gundamConfigCompare
:
Gundam v1.3.3
- Add mirror dial option
- Adding
gundamPlotExtractor
Full changelog of 1.3.2-> 1.3.3
Example for using the new feature mirror dial in the parameter definition config:
- parameterName: "parameterName"
isEnabled: true
dialSetDefinitions:
- dialsType: Spline
applyOnDataSets: [ "DetectorName" ]
dialLeafName: "dialLeafName"
useMirrorDial: true
mirrorLowEdge: -1.
mirrorHighEdge: 1.
Gundam v1.3.2
- Add template parameter definition (possibility to define free parameters without a covariance matrix)
Full changelog of 1.3.1-> 1.3.2
Config example attached bellow. To run the example, unzip and launch:
gundamFitter -c Fitter.yaml -t 6
Gundam v1.3.1
- Table printouts
Gundam v1.3.0
- Added option to set flat prior on specific parameters
- Covariance matrices are now stripped (removing fixed parameter) before inversion
- Minimizer now only handles non fixed parameters (reduce the Hessian dimensionality)