How quanxiny/rcr-featureenvy works
RCR-FeatureEnvy is a research implementation for CG-LSMN feature-envy detection, comparing static contrastive objectives and project-incremental continual learning. The main workflow loads graph-pair data, trains a baseline or contrastive/continual variant, evaluates detection metrics, and aggregates experiment results. Continual streams preserve project/case splits and support replay and distillation; relation-aware losses provide an additional training objective. The README documents experiment orchestration and frozen results, but source excerpts do not expose every runtime connection, so only documented or indexed relationships are shown.
Main entry points: run_contrastive.py Contrastive runner · run_continual.py Continual runner · run_baseline.py Baseline runner · original_adapter.py Dataset adapter
Components 26
Data preparation
- Dataset adapter
src/data/original_adapter.py - Fixed splits
src/data/fixed_splits.py - Metadata builder
src/data/metadata_builder.py - Directed anchors
src/data/directed_anchors.py - Token statistics
src/data/token_statistics.py - Contrastive sampler
src/data/contrastive_sampler.py
Detection model
- CG-LSMN detector
src/models/cglsmn_baseline.py - Pair projection
src/models/contrastive_head.py
Training objectives
- Focal objective
src/losses/original_loss.py - Supervised contrastive
src/losses/supervised_contrastive.py - Local relation contrast
src/losses/local_matching_contrastive.py - Directed relation loss
src/losses/directed_relation_contrastive.py
Continual learning
- Project stream builder
src/continual/stream.py - Continual trainer
src/continual/training.py
Experiments and analysis
- Baseline runner
scripts/run_baseline.py - Contrastive runner
scripts/run_contrastive.py - Continual runner
scripts/run_continual.py - Evaluation metrics
src/utils/metrics.py - Checkpoint writer
src/utils/checkpoint.py - Result aggregation
scripts/aggregate_results.py - Continual aggregation
scripts/aggregate_continual.py - Statistical summaries
scripts/summarize_task2_detection_sci.py - Experiment orchestration
scripts/orchestrate_task2_detection_sci.py
Other components
- Researcher
- CG-LSMN dataset
- Frozen results
Connections 40
- Researcher → Baseline runner: runs baseline
README.md - Researcher → Contrastive runner: runs contrastive
README.md - Researcher → Continual runner: runs continual
README.md - CG-LSMN dataset → Dataset adapter: loads graphs
src/data/original_adapter.py - Dataset adapter → Fixed splits: uses split labels
src/data/original_adapter.py - Metadata builder → Fixed splits: uses fixed splits
src/data/metadata_builder.py - Baseline runner → Dataset adapter: loads data
scripts/run_baseline.py - Baseline runner → CG-LSMN detector: trains detector
scripts/run_baseline.py - Baseline runner → Focal objective: uses objective
scripts/run_baseline.py - Baseline runner → Checkpoint writer: saves checkpoint
scripts/run_baseline.py - Baseline runner → Evaluation metrics: evaluates
scripts/run_baseline.py - Contrastive runner → Dataset adapter: loads data
scripts/run_contrastive.py - Contrastive runner → Fixed splits: selects folds
scripts/run_contrastive.py - Contrastive runner → Directed anchors: builds anchors
scripts/run_contrastive.py - Contrastive runner → Token statistics: uses token stats
scripts/run_contrastive.py - Contrastive runner → Contrastive sampler: samples pairs
scripts/run_contrastive.py - Contrastive runner → CG-LSMN detector: trains detector
scripts/run_contrastive.py - Contrastive runner → Pair projection: uses projection
scripts/run_contrastive.py - Contrastive runner → Focal objective: uses objective
scripts/run_contrastive.py - Contrastive runner → Supervised contrastive: uses objective
scripts/run_contrastive.py - Contrastive runner → Local relation contrast: uses objective
scripts/run_contrastive.py - Contrastive runner → Directed relation loss: uses objective
scripts/run_contrastive.py - Contrastive runner → Checkpoint writer: saves checkpoint
scripts/run_contrastive.py - Contrastive runner → Evaluation metrics: evaluates
scripts/run_contrastive.py - Continual runner → Project stream builder: loads stream
scripts/run_continual.py - Continual runner → Dataset adapter: loads data
scripts/run_continual.py - Continual runner → Directed anchors: uses anchors
scripts/run_continual.py - Continual runner → Token statistics: uses token stats
scripts/run_continual.py - Continual runner → CG-LSMN detector: trains detector
scripts/run_continual.py - Continual runner → Pair projection: uses projection
scripts/run_continual.py - Continual runner → Continual trainer: trains stages
scripts/run_continual.py - Continual runner → Checkpoint writer: saves checkpoint
scripts/run_continual.py - Continual trainer → Directed relation loss: computes loss
src/continual/training.py - Continual trainer → Supervised contrastive: computes loss
src/continual/training.py - Continual trainer → Evaluation metrics: computes metrics
src/continual/training.py - Continual aggregation → Frozen results: writes summaries
README.md - Result aggregation → Frozen results: aggregates runs
README.md - Statistical summaries → Frozen results: summarizes evidence
README.md - Experiment orchestration → Contrastive runner: plans experiments
README.md - Researcher → Experiment orchestration: launches plan
README.md