How sergicasesalonso/pompeu-data-team works
Pompeu is a football-possession inference system for the More than a Hack challenge: it turns match tracking and match context into one H/A/D/X label per frame, then validates and scores that submission. The principal path runs from tracking ingestion through cleaning and feature/event extraction, into a selected model and sequence decoder, and finally to CSV output. The repository also supports video-based ball recovery, external-data and learned experts, and robustness evaluation. Some of these branches are documented or indexed but were not sampled, so their internal links are omitted.
Main entry points: build.py Feature builder · hsmm.py HSMM inference · pipeline.py Prediction pipeline · harness.py Validation harness
Components 28
Inputs and tracking
- Match assets
- Game alignment
games.py - Tracking loaders
pompeu/io/atd.py - Tracking contract
pompeu/io/contract.py - Tracking cleaning
pompeu/clean/__init__.py - Repository paths
pompeu/paths.py
Possession inference
- Motion features
pompeu/features/kinematics.py - Feature builder
pompeu/features/build.py - Match context
pompeu/features/context.py - Touch events
pompeu/events/touches.py - Dead-ball detection
pompeu/events/deadball.py - Rule model
pompeu/models/rules.py - HSMM inference
pompeu/models/hsmm.py - Viterbi decoder
pompeu/decode/viterbi.py - Prediction pipeline
pompeu/pipeline.py - Submission CSV
scripts/make_submission.py
Experts and vision
- Broadcast video
- Video ball fusion
pompeu/vision/fuse.py - External matches
- External examples
pompeu/external/examples.py - Data degradation
pompeu/external/degrade.py - Synthetic tags
pompeu/external/synthetic_tags.py - TCN expert
pompeu/experimental/tcn.py
Validation and scoring
- Stress validation
pompeu/validate/stress.py - Validation harness
pompeu/validate/harness.py - Public ground truth
pompeu/io/gt.py - Possession metrics
evaluation/possession_metrics.py - Evaluation runner
evaluation/run_evaluation.py
Connections 36
- Match assets → Tracking loaders: supplies data
- Game alignment → Tracking contract: defines alignment
pompeu/io/contract.py - Tracking loaders → Tracking contract: loads tracking
pompeu/io/atd.py - Tracking contract → Tracking cleaning: provides tracking
pompeu/clean/__init__.py - Prediction pipeline → Tracking cleaning: cleans tracking
pompeu/pipeline.py - Tracking cleaning → Motion features: uses motion features
- Feature builder → Motion features: computes motion
pompeu/features/build.py - Feature builder → Match context: builds context
pompeu/features/build.py - Match context → Game alignment: reads match constants
pompeu/features/context.py - Touch events → Motion features: uses motion
pompeu/events/touches.py - Dead-ball detection → Motion features: uses motion
pompeu/events/deadball.py - Dead-ball detection → Feature builder: uses features
pompeu/events/deadball.py - Rule model → Dead-ball detection: uses events
pompeu/models/rules.py - Rule model → Touch events: uses events
pompeu/models/rules.py - HSMM inference → Feature builder: builds evidence
pompeu/models/hsmm.py - HSMM inference → Dead-ball detection: uses detection
pompeu/models/hsmm.py - HSMM inference → Touch events: uses events
pompeu/models/hsmm.py - HSMM inference → Rule model: uses rule labels
pompeu/models/hsmm.py - HSMM inference → Viterbi decoder: decodes states
pompeu/models/hsmm.py - Prediction pipeline → Rule model: dispatches method
pompeu/pipeline.py - Prediction pipeline → HSMM inference: dispatches method
pompeu/pipeline.py - Prediction pipeline → TCN expert: dispatches method
pompeu/pipeline.py - Prediction pipeline → Submission CSV: provides predictions
- Broadcast video → Video ball fusion: provides video
- External matches → External examples: provides matches
- External examples → Data degradation: degrades examples
pompeu/external/examples.py - External examples → Feature builder: builds features
pompeu/external/examples.py - Synthetic tags → Feature builder: injects context
pompeu/external/synthetic_tags.py - TCN expert → Feature builder: builds features
pompeu/experimental/tcn.py - Validation harness → Prediction pipeline: runs predictions
pompeu/validate/harness.py - Validation harness → Stress validation: runs stress checks
pompeu/validate/harness.py - Validation harness → Public ground truth: loads truth
pompeu/validate/harness.py - Validation harness → Possession metrics: computes scores
pompeu/validate/harness.py - Evaluation runner → Possession metrics: scores submissions
evaluation/run_evaluation.py - Repository paths → Tracking loaders: sets data paths
pompeu/io/atd.py - Repository paths → External examples: sets data paths
pompeu/external/examples.py