How ajharshal45/just_to_learn works
This repository is a research benchmark and interactive platform for comparing conventional and SHAP-guided oversampling on imbalanced tabular classification. The main workflow is dataset selection or upload, leak-free experiment execution, oversampling and model evaluation, then inspection of metrics, explanations, fidelity, and persisted results. It also includes a public-facing research site and a standalone results simulator. The graph combines documented protocol stages with sampled API/UI wiring; several internal calls are supported only by the source index, so unsampled links are omitted.
Main entry points: experiment_service.py Experiment service · runner.py Leak-free runner · main.py REST endpoints · oversampling Oversampling methods
Components 25
Research interfaces
- Platform pages
frontend/src/pages - API client
frontend/src/services/api.ts - API contracts
frontend/src/types/api.ts - Shared UI
frontend/src/components/common - Research website
website/src/App.jsx - Website simulator
website/src/components/Simulator.jsx - Simulator data
website/src/data/simulatorData.js - Standalone simulator
simulator-app/src/App.jsx - Simulator data
simulator-app/src/data/simulatorData.js
Platform API
- REST endpoints
src/api/main.py - Dataset management
src/api/dataset_service.py - Experiment service
src/api/experiment_service.py
Experiment workflow
- Dataset loaders
src/pipeline/datasets.py - Benchmark config
src/experiment/config.py - Leak-free runner
src/experiment/runner.py - Fold preprocessing
src/experiment/preprocessing.py
ML methods
- Oversampling methods
src/oversampling - SHAP weighting
src/oversampling/base.py - Classifiers
src/pipeline/models.py - Metrics and stability
src/evaluation - Synthetic fidelity
src/evaluation/fidelity.py
Results and persistence
- Run inspection
src/experiment/inspection.py - SQLite results
src/api/database.py
Other components
- Researcher
- Dataset files
Connections 33
- Researcher → Platform pages: uses
- Platform pages → API client: requests
frontend/src/pages/DatasetExplorer.tsx - Platform pages → Shared UI: renders
frontend/src/pages/DatasetExplorer.tsx - API client → API contracts: uses types
frontend/src/services/api.ts - API client → REST endpoints: calls REST
frontend/src/services/api.ts - Researcher → Research website: explores research
website/src/App.jsx - Research website → Website simulator: renders
website/src/App.jsx - Website simulator → Simulator data: reads data
website/src/components/Simulator.jsx - Researcher → Standalone simulator: runs simulation
simulator-app/src/App.jsx - Standalone simulator → Simulator data: reads data
simulator-app/src/App.jsx - REST endpoints → Dataset management: routes dataset requests
src/api/main.py - REST endpoints → Experiment service: routes experiment requests
src/api/main.py - REST endpoints → SQLite results: reads and writes
src/api/main.py - Dataset management → Dataset loaders: loads benchmarks
src/api/dataset_service.py - Dataset management → SQLite results: stores uploads
src/api/dataset_service.py - Experiment service → Dataset loaders: loads experiment data
src/api/experiment_service.py - Experiment service → Oversampling methods: selects sampler
src/api/experiment_service.py - Experiment service → Classifiers: selects classifier
src/api/experiment_service.py - Experiment service → Synthetic fidelity: measures fidelity
src/api/experiment_service.py - Experiment service → Leak-free runner: runs cross-validation
src/api/experiment_service.py - Experiment service → SQLite results: persists run
src/api/experiment_service.py - Benchmark config → Oversampling methods: configures methods
src/experiment/config.py - Benchmark config → Classifiers: configures models
src/experiment/config.py - Leak-free runner → Benchmark config: uses settings
src/experiment/runner.py - Leak-free runner → Fold preprocessing: fits in folds
src/experiment/runner.py - Leak-free runner → Metrics and stability: computes metrics
src/experiment/runner.py - Leak-free runner → Oversampling methods: resamples training folds
- Leak-free runner → Classifiers: fits classifiers
- Oversampling methods → SHAP weighting: uses attributions
src/oversampling/__init__.py - Experiment service → Run inspection: produces diagnostics
src/api/experiment_service.py - Run inspection → Synthetic fidelity: analyzes samples
src/experiment/inspection.py - Dataset files → Dataset loaders: provides datasets
- REST endpoints → Run inspection: serves explanations