How mohithbs-stack/resolvent works
This repository provides an end-to-end business entity resolution challenge solution: it ingests three noisy business-record sources, generates candidate pairs, scores likely matches, and packages matches plus the candidate audit trail. It also supports held-out F0.5 evaluation and a dashboard-data export path. The sampled orchestration excerpt is incomplete, so links between pipeline stages reflect the documented stage sequence; dashboard rendering details are not evident from the sampled source.
Main entry points: run_pipeline.py Pipeline orchestrator · candidate_gen.py Candidate blocking · similarity.py Similarity features · predict.py Score & threshold
Components 14
Resolution Workflow
- Pipeline orchestrator
code/business_entity_resolution/run_pipeline.py - Load & normalize
code/business_entity_resolution/src/ingest/load_normalize.py - Candidate blocking
code/business_entity_resolution/src/blocking/candidate_gen.py
Matching Model
- Similarity features
code/business_entity_resolution/src/features/similarity.py - Train matcher
code/business_entity_resolution/src/matching/train.py - LightGBM model
models/lgbm_matcher.txt - Score & threshold
code/business_entity_resolution/src/matching/predict.py
Evaluation & Reporting
- F0.5 validation
code/business_entity_resolution/src/evaluate/score_f05.py - Dashboard data export
code/business_entity_resolution/export_dashboard_data.py - Dashboard
dashboard/index.html
Submission Outputs
- Package outputs
code/business_entity_resolution/src/submit/package.py - Submission TSVs
Other components
- Challenge operator
- Source TSVs
Connections 19
- Challenge operator → Pipeline orchestrator: runs
- Source TSVs → Pipeline orchestrator: supplies
- Pipeline orchestrator → Load & normalize: loads data
- Pipeline orchestrator → Candidate blocking: generates candidates
- Pipeline orchestrator → Similarity features: builds features
- Pipeline orchestrator → Train matcher: trains matcher
- Pipeline orchestrator → Score & threshold: runs inference
- Load & normalize → Candidate blocking: normalized records
- Candidate blocking → Similarity features: candidate pairs
- Similarity features → Train matcher: training features
- Train matcher → LightGBM model: saves model
- Similarity features → Score & threshold: inference features
- LightGBM model → Score & threshold: scores with
- Score & threshold → Package outputs: match sets
- Candidate blocking → Package outputs: candidate sets
- Package outputs → Submission TSVs: writes TSVs
- F0.5 validation → Dashboard data export: provides scoring
- LightGBM model → Dashboard data export: feature importances
- Dashboard data export → Dashboard: exports dashboard data