v0.10.13

Fixed

• Docker: Bundle OpT0Finder v1.0.0 and the ICARUS PhotonLibrary so likelihood flash matching works in the published image.

Full Changelog: v0.10.12...v0.10.13

v0.10.12

Fixed

• Inference: Run model-less inference jobs once so reader/build/post/writer workflows execute without requiring a model block.

Full Changelog: v0.10.11...v0.10.12

v0.10.11

Added

• Configuration: Add a spine-config command and bin/config.py proxy to load complex SPINE configs, dump the resolved YAML, and compare resolved configs.

Changed

• HDF5 Writer: Refactor output file handling, split-output naming, dataset creation, and append/opening logic for clearer and more consistent writer behavior.
• CSV Writer: Align writer naming and type handling with the cleaned-up HDF5 writer behavior.

Fixed

• Docker: Build h5py against the system HDF5 library used by LArCV to avoid HDF5 ABI mismatches when importing LArCV and h5py in the same process.
• Full Chain: Handle empty fragment-list group indexes without failing on an empty maximum reduction.

Full Changelog: v0.10.10...v0.10.11

v0.10.10

Changed

• Package import: Lazy-load Driver from the top-level spine package so lightweight imports such as spine.config do not load the full driver stack.

Fixed

• Tests: Avoid mocking unavailable PyTorch with a sys.modules sentinel in conditional import tests.

Full Changelog: v0.10.9...v0.10.10

v0.10.9

Added

• Configuration: Support for environment variable expansion in configuration files
• I/O: Remote XRootD input path support for distributed file access
• GrapPA: Configurable feature outputs for GrapPA module (#123)
• Inference: Support for lists of inference weight paths for multi-model workflows (#122)

Changed

• Docker: Enhanced CI/CD with Buildx and Docker layer caching
• Truth Matching: Track original point indexes for improved truth matching (#121)
• Documentation: Clarified Docker usage documentation
• Validation: Tightened module weight path validation (#122)

Full Changelog: v0.10.8...v0.10.9

v0.10.8

Added

• Docker Containerization: Complete Docker infrastructure for production deployments
  • Full ML stack with PyTorch 2.5.1, MinkowskiEngine v0.5.4, torch-geometric, ROOT, and LArCV2
  • Ubuntu 22.04 base with CUDA 12.1 toolkit (perfect version match with PyTorch)
  • XRootD client with SciTokens support for dCache streaming with token authentication
  • Multi-GPU architecture support: V100, A100, H100/H200, RTX 20xx/30xx/40xx (compute 7.0-9.0)
  • Automated GitHub Actions workflow for container builds and publishing to GHCR
  • Comprehensive documentation with Apptainer/Singularity usage examples
  • Build script for local development and testing

Changed

• Dependencies: Removed torch-sparse dependency (no longer required)
• Documentation: Updated all Singularity references to Apptainer (current standard)
• Sphinx: Removed torch-sparse from autodoc mock imports
• Docker: Local Docker builds now force-refresh the base image with --pull
• Feature/shower energy by @francois-drielsma in #120

Fixed

• NumPy 2: Avoid coercing EventSparseTensor3D lists into NumPy arrays in Sparse3DParser

Full Changelog: v0.10.6...v0.10.8

v0.10.6

Changed

• CSV Writer: Significantly improved CSV writer performance for analysis scripts (#119)
• Multi-node training: Fixed multi-node distributed training support (#118)

Full Changelog: v0.10.5...v0.10.6

v0.10.5

Fixed

• Visualization: Fixed raw drawing to behave correctly for truth data
• Track analysis: Fixed bug in track completeness algorithm

Full Changelog: v0.10.4...v0.10.5

v0.10.4

Added

• CLI: Added --entry-list and --skip-entry-list arguments for easy entry filtering
• Multi-node training: Enhanced main.py to support multi-node distributed training

Fixed

• File downloads: Added file locking to prevent race conditions when multiple jobs download files concurrently
• CLI: Guard against null loader/reader to prevent crashes
• Checkpointing: Fixed epoch-based weight checkpointing bug
• Vertexer: Proper singularity check in vertex computation

Full Changelog: v0.10.3...v0.10.4

v0.10.3

Added

• Visualization: Added support to draw _sum attributes in Drawer

Fixed

• Data objects: Fixed default typing issue in crt_times/crt_scores attributes of Interaction objects
• Visualization: Do not make a single attribute in the drawer be the default color scale

Changed

• Visualization: Use asdict to get object properties in Drawer (more complete)
• Config loading: Added explicit message when SPINE_CONFIG_PATH is not set and config include is not found

Full Changelog: v0.10.2...v0.10.3