How jay4ignite/ml-cornerstone works
This repository is a config-driven loan default-risk estimator with a scikit-learn training pipeline and an in-process Streamlit application—not an approval engine or HTTP service. The training workflow acquires or generates data, standardizes and validates it, splits and trains candidates, selects and calibrates a model, evaluates and explains it, then persists artifacts. In the app, an applicant submits inputs for validation and prediction; the result presents probability, score, band, limitations, and explanatory context. README documents additional analytics and governance pages; unsampled implementations are represented at their listed paths.
Main entry points: run.py Training workflow · streamlit_app.py Streamlit app · service.py Model service · assessment.py Applicant assessment
Components 34
Data preparation
- Configuration
loan_risk/config.py - Feature contract
loan_risk/schema.py - Dataset acquisition
loan_risk/data/download.py - Synthetic data
loan_risk/data/synthetic.py - Data ingestion
loan_risk/data/ingestion.py - Data validation
loan_risk/data/validate.py - Dataset manifest
loan_risk/data/manifest.py
Model training
- Training workflow
loan_risk/pipeline/run.py - Data splitting
loan_risk/pipeline/split.py - Feature preprocessing
loan_risk/pipeline/preprocess.py - Candidate models
loan_risk/pipeline/models.py - Model tuning
loan_risk/pipeline/tuning.py - Probability calibration
loan_risk/pipeline/calibrate.py - Model evaluation
loan_risk/pipeline/evaluate.py - Fairness diagnostics
loan_risk/pipeline/fairness.py - Model explanations
loan_risk/pipeline/explain.py - Model artifacts
loan_risk/pipeline/artifacts.py
Prediction serving
- Model service
loan_risk/serving/service.py - Input validation
loan_risk/serving/validation.py - Risk mapping
loan_risk/pipeline/risk_score.py
Assessment interface
- Streamlit app
streamlit_app.py - Applicant assessment
ui/pages/assessment.py - Session prediction store
ui/state.py - Prediction results
ui/pages/result.py - Local explanations
ui/pages/explainability.py - Shared UI helpers
ui/helpers.py
Analytics and governance
- Performance analytics
ui/pages/performance.py - Analytics dashboard
ui/pages/analytics.py - Model comparison
ui/pages/model_selection.py - Dataset quality
ui/pages/dataset_quality.py - Feature reference
ui/pages/feature_reference.py - Methodology
ui/pages/methodology.py - Model information
ui/pages/model_info.py
Other components
- Applicant
Connections 42
- Configuration → Dataset acquisition: configures
loan_risk/data/download.py - Configuration → Data ingestion: selects source
loan_risk/data/ingestion.py - Feature contract → Data ingestion: defines columns
loan_risk/data/ingestion.py - Training workflow → Dataset acquisition: acquires data
loan_risk/pipeline/run.py - Training workflow → Synthetic data: generates fixture
loan_risk/pipeline/run.py - Training workflow → Data ingestion: loads and standardizes
loan_risk/pipeline/run.py - Training workflow → Data validation: validates data
loan_risk/pipeline/run.py - Training workflow → Dataset manifest: builds manifest
loan_risk/pipeline/run.py - Training workflow → Data splitting: splits data
loan_risk/pipeline/run.py - Training workflow → Candidate models: builds candidates
loan_risk/pipeline/run.py - Training workflow → Model tuning: tunes candidates
loan_risk/pipeline/run.py - Training workflow → Probability calibration: calibrates selected
loan_risk/pipeline/run.py - Training workflow → Model evaluation: evaluates test set
loan_risk/pipeline/run.py - Training workflow → Fairness diagnostics: runs diagnostics
loan_risk/pipeline/run.py - Training workflow → Model explanations: computes explanations
loan_risk/pipeline/run.py - Training workflow → Model artifacts: persists artifacts
loan_risk/pipeline/run.py - Feature contract → Feature preprocessing: defines features
loan_risk/pipeline/preprocess.py - Feature preprocessing → Candidate models: transforms inputs
loan_risk/pipeline/models.py - Applicant → Applicant assessment: enters application
README.md - Streamlit app → Applicant assessment: routes page
streamlit_app.py - Applicant assessment → Input validation: validates payload
ui/pages/assessment.py - Applicant assessment → Model service: requests prediction
ui/pages/assessment.py - Model service → Model artifacts: loads model
loan_risk/serving/service.py - Model service → Feature contract: builds model frame
loan_risk/serving/service.py - Model service → Risk mapping: maps probability
loan_risk/serving/service.py - Model service → Model explanations: computes local explanation
loan_risk/serving/service.py - Applicant assessment → Session prediction store: saves prediction
ui/pages/assessment.py - Prediction results → Session prediction store: reads prediction
ui/pages/result.py - Local explanations → Session prediction store: reads prediction
ui/pages/explainability.py - Streamlit app → Prediction results: routes page
streamlit_app.py - Streamlit app → Local explanations: routes page
streamlit_app.py - Streamlit app → Performance analytics: routes page
streamlit_app.py - Streamlit app → Dataset quality: routes page
streamlit_app.py - Streamlit app → Feature reference: routes page
streamlit_app.py - Streamlit app → Methodology: routes page
streamlit_app.py - Streamlit app → Model information: routes page
streamlit_app.py - Streamlit app → Model comparison: routes page
streamlit_app.py - Streamlit app → Analytics dashboard: routes page
streamlit_app.py - Shared UI helpers → Model service: provides cached service
ui/helpers.py - Performance analytics → Shared UI helpers: uses rendering helpers
ui/pages/performance.py - Dataset quality → Shared UI helpers: uses rendering helpers
ui/pages/dataset_quality.py - Feature reference → Feature contract: references feature contract