How myradvd/finfrauddetml-research works
Fraudlab is a reproducible financial-fraud ranking pipeline. Raw training and test CSVs enter through the CLI: data audit and shift analysis inform feature construction and model selection; training then fits the selected model, scores test rows, and writes a submission. The README also documents a standalone prediction notebook and exploratory model/shift experiments. Source excerpts support the main pipeline calls; notebook and experiment connections are documented in the README, not verified from their implementation here.
Main entry points: select.py Model selection · train.py Model training · __main__.py Pipeline CLI · data.py Data loading and cleaning
Components 19
Pipeline workflow
- Pipeline CLI
fraudlab/__main__.py - Schema and settings
fraudlab/config.py - Data loading and cleaning
fraudlab/data.py - Model training
fraudlab/train.py
Features and modeling
- Feature engineering
fraudlab/features.py - Model selection
fraudlab/select.py - Model zoo
fraudlab/models.py
Analysis and evaluation
- Audit and shift analysis
fraudlab/eda.py - Analysis plots
fraudlab/plots.py - CV and metrics
fraudlab/evaluate.py - Selection artifacts
results/model_selection - Model and shift experiments
notebooks - Upload scores
results/leaderboard.csv
Prediction delivery
- Fitted model bundle
final_notebook/model.pkl - Test scoring and submission
fraudlab/train.py - Prediction submission
results/final - Standalone prediction notebook
final_notebook/predict_catboost.ipynb
Other components
- Pipeline operator
- Train and test CSVs
Connections 28
- Pipeline operator → Pipeline CLI: runs stages
README.md - Train and test CSVs → Data loading and cleaning: loads CSVs
fraudlab/data.py - Pipeline CLI → Audit and shift analysis: dispatches
fraudlab/__main__.py - Pipeline CLI → Model selection: dispatches
fraudlab/__main__.py - Pipeline CLI → Model training: dispatches
fraudlab/__main__.py - Pipeline CLI → Test scoring and submission: dispatches
fraudlab/__main__.py - Audit and shift analysis → Data loading and cleaning: loads and sanitises
fraudlab/eda.py - Audit and shift analysis → Feature engineering: engineers features
fraudlab/eda.py - Audit and shift analysis → Analysis plots: generates figures
fraudlab/eda.py - Model selection → Data loading and cleaning: loads training data
fraudlab/select.py - Model selection → Feature engineering: builds features
fraudlab/select.py - Model selection → CV and metrics: cross-validates
fraudlab/select.py - Model selection → Model zoo: scores candidates
fraudlab/select.py - Model selection → Analysis plots: plots comparison
fraudlab/select.py - Model selection → Upload scores: reads public scores
fraudlab/select.py - Model selection → Selection artifacts: writes selection
fraudlab/select.py - Model training → Selection artifacts: reads chosen model
fraudlab/train.py - Model training → Data loading and cleaning: loads training data
fraudlab/train.py - Model training → Feature engineering: builds features
fraudlab/train.py - Model training → Model zoo: fits selected learners
fraudlab/train.py - Model training → Fitted model bundle: saves model
fraudlab/train.py - Test scoring and submission → Fitted model bundle: loads model
fraudlab/train.py - Test scoring and submission → Data loading and cleaning: loads test data
fraudlab/train.py - Test scoring and submission → Feature engineering: builds features
fraudlab/train.py - Test scoring and submission → Prediction submission: writes predictions
fraudlab/train.py - Standalone prediction notebook → Fitted model bundle: loads model
README.md - Standalone prediction notebook → Prediction submission: produces submission
README.md - Model and shift experiments → Upload scores: records uploads
README.md