v2.9.2.2

Patch to analysis w.r.t. the previous release:

• Added timing for the CSV writing
• Correctly store primary particle count in CSV
• Bug fix in uncertainty evaluation for efficiency/purity

v2.9.2.1

Patches/new features in mlreco:

• Replaced scipy.optimize.fsolve by scipy.optimize.brentq in csda_ke_lar function to get rid of failure cases
• Added more diagnostic tools for gSPICE
• Added new loss to gSPICE

Important patches in analysis:

• The number of children associated with a Particle is now loaded properly from file
• Bug when running mlreco + analysis in one job fixed
• Bayes efficiency/purity uncertainty estimator
• Add topology selection tool

v2.9.2

New features and bug fixes in mlreco:

• Brand new calibration package under mlreco.utils, soon to be integrated in the full chain
• Multiple corner-case bug fixes in cluster_direction algorithm
• Added an option to set the relative and absolute precision of the Bethe Bloch integration in the energy_loss module
• Fixed tracking module functions slowdown. np.cov more reliable in numba than np.dot, where relevant

Bug fixes in analysis:

• Can now easily load 2x2_mr3 or 2x2_mr4 geometries by name trivially
• Substract rest mass from true KE value when using energy_init

v2.9.1.5

Fix bug in mlreco:

• Multi-volume forward (for ICARUS) was broken at the unwrapping stage. Now fixed.

Small enhancement in analysis:

• TruthParticle object now load gen_id from larcv.Particle objects, when available
• Added np.clip(X, -1, 1) no cosine of angles in MCS to avoid floating point precision issues

v2.9.1.4

Updates to mlreco:

• Finalized mlreco.utils.mcs, now works well and fast (some fine tuning still needed)
• Add option in HDF5Reader to read in an entry by [run, event] numbers
• Add parser to read in new larcv.Trigger objects
• Fixed mlreco.utils.numba_local.principal_components to run much quicker (using np.cov)

New features and bug fixes in analysis tools:

• truth_momentum and truth_start_dir attributes of TruthParticle objects set properly

v2.9.1.3

Upgrades/fixes to HDF5Writer/HDF5Reader:

• Added option to load a specific [run, event] pair using the HDF5Reader
• Handle uneven batch size more gracefully: no more fails if batch_size changes for the last batch (total number of events not a multiple of the batch_size

Bug fixes in analysis tools:

• Make sure that the KE cuts are reset before applying new ones in the adjust_interaction_topology post-processor
• Load truth_momentum and truth_start_dir properly from analyzed HDF5 files

v2.9.1.2

Updates to the analysis tools:

• Added reconstruct_mcs_energy post-processor (needs to be optimized)
• Added adjust_interaction_topology post-processor to update the topology of interactions based on KE cuts
• Removed reconstruct_momentum which is now a simple property of TruthParticle objects by default
• Fixed bug that would not allow the post-processors to run without matcha installed
• Fixed bug that would make it impossible to initialize Geometry without source/opdets files
• Fixed bug when running CSDA/Vertex post-processors on TruthParticle objects

v2.9.1.1

A few essential bug fixes:

• Fixed issue with neutrino information storage/loading
• Fixed bug with flash matching when some space point have a negative charge (can happen when using collection plane charge only and the collection plane is missing when forming the space point)
• Fixed bug with the reconstruct_momentum post-processor when the KE is negative

v2.9.1

Updates to mlreco:

• Expended the Geometry class to include the location of the optical detectors
• Refactored the check_containment function of the Geometry class so that the volume to check against can be selected once and applied many times (saves time). It now also includes an option to apply a different containment cut at the cathode
• Fixed a small offset between the default ICARUS Geometry and the correct geometry

Updates to the analysis tools:

• Complete refactoring of the post-processors. They are now classes which inherit from a common PostProcessor parent class and are initialized once at the start of the analysis tool manager process. Saves time and allows for more flexibility
• Completely rewrote the OpT0Finder interface. This is associated with both an execution time reduction and an increase in accuracy thanks to a bug fix in the x shift applied to modules that are not the reference module
• Added a new barycenter-method flash matching post-processor
• Added option to include a fudge factor to the shower calorimetric energy reconstruction
• Added neutrino truth attributes to the TruthInteraction object

Some of the above changes will break older analysis tools configurations, particularly when it comes to the flash matching and containment post-processors.

v2.9.0.5

Bug fix in mlreco:

• Always store the selected rescaled charge to input_rescaled only