Gundam v1.2.8

• Fixed prior value override when specified in parameter config
• Added app "gundamConfigUnfolder" which forward every sub-config file, and write the global config in a single json file
• gundamFitter now writes infos about the runtime (gundam version, command line args, unfolded config) in the output rootfile

Full changelog of 1.2.7 -> 1.2.8

Gundam v1.2.7

• Fixing buggy progressbar
• Don't scan fixed parameters
• Swapping header order (to better identify missing includes)

Full changelog of 1.2.6 -> 1.2.7

Gundam v1.2.6

• Fitter parameters can now be linearly transformed in a normalized scale: centered at 0, std dev at 1.
• Fixed conversion of eigen <-> original fit phase space
• Now computing chi2 in the eigen phase space if enabled (faster)
• Can specify the maxNbEigenParameters (instead of the max eigen fraction)
• Add option fixGhostEigenParmetersAfterFirstRejected
• Decomposition of the post-fit error matrix / output Hessian to plot the contribution of each Hessian eigen parameter of each fit parameter
• Add fillStyle option for plot generator dictionnary
• Limit the number of entries in the legend of each generated plot (unspecified plots are set to gray)
• ENABLE_DEV_MODE cmake option to provide the line number/file of each printout

1.2.5 -> 1.2.6

Gundam v1.2.5

• Extracting post-fit eigen composition of the Hessian
• Changed post-fit errors plots to see the prior value
• Fixed throw correlated parameters for Asimov and Mc before fit
• Adding custom binning definition for plot generator

1.2.4 -> 1.2.5

Gundam v1.2.4

Erratum for 1.2.3:

• Shift normalization weight when the prior is not 1 -> NO!

Gundam v1.2.3

• Add fake data weight
• Shift normalization weight when the prior is not 1
• Decomposition of the output Hessian is now optionnal

1.2.2 -> 1.2.3

Gundam v1.2.2

• Fixing event by event splines
• Adding graph dials
• "additionnalLeavesStorage" option in dataset
• Faster event loader
• removing mutex lock -> using atomic instead
• faster LLH calculation (var buffer are defined as members)
• Printing initial chi2
• Disabling default chi2 scan

1.2.1 -> 1.2.2

Gundam v1.2.1

Minor revision:

• Mostly printouts and minor debugs
• Scan steps now set into the fitter engine

1.2.0 -> 1.2.1

Gundam v1.2.0

Several improvements and stability features were implemented:

• OpenMP now fully optional
• Better Yaml-CPP finder
• Install instruction for CCLyon
• Fixed memory leak in plot generator
• gundamFitter: adding options --one-sigma and --scan
• Can now load data from external files
• Writing sample events in dedicated TTrees
• Optimized dial indexing
• A lot less RAM is required to fit with big data sets (only keeping the necessary variables in memory)
• Fixed several memory issues appearing while running with parallel threads
• Faster plot generator (using a bin cache)
• Displaying sample breakdown when loaded
• Fixed eigen decomposition
• Spline dials now store shared_ptr of TSpline3 (no more memory leaks)
• Event-by-event splines
• Auto write bad splines when preventing negative response
• Adding config option to scale parameter step
• Other various improvements...

1.1.0 -> 1.2.0

Gundam v1.1.0

Several improvements and stability features were implemented:

• Fixed Yaml-cpp cmake find_package
• Less RAM is required to run the fitter (dial indexing is now done within the data loader, leaving unncessary branches out of memory)
• Event-by-event splines
• Simplified data reader
• Explicit minuit returned status (convergence state value + meaning)
• PhysicsEvent now handles clonning of pointer leaves
• Renamed packages

1.0.0 -> 1.1.0