How heidelberg-hepml/lorentz-gatr works
L-GATr is a Lorentz-equivariant geometric-algebra transformer library, accompanied by high-energy-physics experiments for amplitude regression, jet tagging, and event generation. The user workflow is to configure and launch an experiment, prepare its task-specific data, train or evaluate a model, and inspect logged results; the README also describes using the network and geometric-object interfaces directly. The graph includes the principal model-building layers and the documented experiment branches. Some subsystem links rely on the README or SOURCE INDEX references because their implementations were not excerpted.
Main entry points: base_experiment.py Experiment runtime · gatr_block.py Transformer block · run.py Experiment launcher · experiment.py Jet tagging
Components 28
Experiment entry
- Experiment launcher
run.py
Physics experiments
- Amplitude regression
experiments/amplitudes/experiment.py - Amplitude data
experiments/amplitudes/dataset.py - Jet tagging
experiments/tagging/experiment.py - Tagging data
experiments/tagging/dataset.py - Jet embedding
experiments/tagging/embedding.py - MiniWeaver ingestion
experiments/tagging/miniweaver/dataset.py - Event generation
experiments/eventgen/experiment.py - Event data
experiments/eventgen/dataset.py - Event transforms
experiments/eventgen/transforms.py
Geometric representation
- GA interfaces
gatr/interface/vector.py - Symmetry references
gatr/interface/spurions.py
Equivariant network
- GATr network
gatr/nets/gatr.py - Axial GATr
gatr/nets/axial_gatr.py - Transformer block
gatr/layers/gatr_block.py - Attention configuration
gatr/layers/attention/config.py - Attention primitives
gatr/primitives/attention.py - Geometric MLP
gatr/layers/mlp/mlp.py - Geometric bilinears
gatr/layers/mlp/geometric_bilinears.py - Equivariant linear layers
gatr/layers/linear.py - Linear primitives
gatr/primitives/linear.py - Tensor utilities
gatr/utils/einsum.py
Training and tracking
- Experiment runtime
experiments/base_experiment.py - Logging
experiments/logger.py - Run tracking
experiments/mlflow.py
Other components
- Researcher
- Dataset host
- Tracking store
Connections 34
- Researcher → Experiment launcher: launches
README.md - Experiment launcher → Amplitude regression: dispatches
run.py - Experiment launcher → Jet tagging: dispatches
run.py - Experiment launcher → Event generation: dispatches
run.py - Dataset host → Amplitude data: provides data
README.md - Dataset host → Tagging data: provides data
README.md - Dataset host → Event data: provides data
README.md - Amplitude regression → Amplitude data: loads
experiments/amplitudes/experiment.py - Jet tagging → Tagging data: uses
experiments/tagging/experiment.py - Jet tagging → Jet embedding: uses
experiments/tagging/experiment.py - Jet embedding → Tagging data: imports from
experiments/tagging/embedding.py - MiniWeaver ingestion → Tagging data: supports ingestion
- Event generation → Event data: loads
experiments/eventgen/experiment.py - Event generation → Event transforms: uses
- Amplitude regression → Experiment runtime: extends
experiments/amplitudes/experiment.py - Jet tagging → Experiment runtime: extends
experiments/tagging/experiment.py - Event generation → Experiment runtime: extends
experiments/eventgen/experiment.py - Experiment runtime → Logging: logs through
experiments/base_experiment.py - Experiment runtime → Run tracking: tracks runs
experiments/base_experiment.py - Run tracking → Logging: uses
experiments/mlflow.py - Run tracking → Tracking store: persists to
README.md - Jet embedding → GA interfaces: embeds vectors
experiments/tagging/embedding.py - Jet embedding → Symmetry references: embeds references
experiments/tagging/embedding.py - GA interfaces → Symmetry references: provides exports
gatr/interface/__init__.py - GATr network → Transformer block: builds from
gatr/nets/gatr.py - Axial GATr → Transformer block: builds from
gatr/nets/axial_gatr.py - Transformer block → Attention configuration: configures
gatr/layers/gatr_block.py - Transformer block → Geometric MLP: uses
gatr/layers/gatr_block.py - Geometric MLP → Geometric bilinears: uses
gatr/layers/mlp/mlp.py - Geometric MLP → Equivariant linear layers: uses
gatr/layers/mlp/mlp.py - Equivariant linear layers → Linear primitives: calls
gatr/layers/linear.py - Linear primitives → Tensor utilities: uses
gatr/primitives/linear.py - Transformer block → Attention primitives: attention operations
- Experiment runtime → GATr network: instantiates configured model