Resources
Everything behind the project
The deck, the reports, every notebook, the data sources and the saved models, all in one place.
Deck
The CEROP story in 14 slides
Readable in about five minutes without a speaker. View in the browser, or download the PDF from the last slide.
Final report · PDFThe full write-up
Every method, table and figure as submitted at the end of the course.
Deck PDF: CEROP_Deck.pdf
Notebooks
Copied from the course folder with their outputs intact. The only edits are redacted local file paths, listed in docs/redactions.md.
| Notebook | What it does | Author | Open |
|---|---|---|---|
01_eda_combined | Week 3 exploration of both datasets | Arthunya (financial), Aswin (operations) | GitHub · nbviewer |
02a_financial_cleaning | Sentinel values, capping, composite features | Arthunya | GitHub · nbviewer |
02b_operations_cleaning | Leakage columns, outliers, date features | Aswin | GitHub · nbviewer |
03_pipeline_integration | Both layers in one split-then-fit pipeline, GSCPI merge, checks | Divya | GitHub · nbviewer |
04a_financial_model_selection | Logistic Regression vs Random Forest vs XGBoost | Ankit | GitHub · nbviewer |
04b_operations_model_selection | Logistic Regression vs linear SVM vs Random Forest | Divya | GitHub · nbviewer |
05a_financial_finalization | Tuning, threshold checks, bootstrap | Ankit | GitHub · nbviewer |
05b_operations_tuning | Random Forest search and threshold calibration | Divya | GitHub · nbviewer |
06_live_demo_scoring_and_shap | The demo scripts, run end to end; the profit-ratio scale check | Arthunya | GitHub · nbviewer |
07_export_web_artifacts | Scores, SHAP cases and the what-if model behind this site | Arthunya | GitHub · nbviewer |
Reports
The weekly group reports and the Final Report, exported to PDF from the submitted documents.
| Week | Report |
|---|---|
| 2 | Frame the problem and proposal |
| 3 | Getting and exploring the data |
| 4 | Preparing the data |
| 5 | Model selection and shortlisting |
| 6 | Finalizing the models |
| Final | Final Report |
Datasets
Raw data isn't stored in the repo. data/README.md says where to download each file and where to put it.
| Dataset | Source | License |
|---|---|---|
| Taiwanese Bankruptcy Prediction | UCI Machine Learning Repository (doi:10.24432/C5004D) | CC BY 4.0 |
| DataCo Smart Supply Chain for Big Data Analysis | Mendeley Data, Constante, Silva & Pereira (doi:10.17632/8gx2fvg2k6.5) | CC BY 4.0 |
| Global Supply Chain Pressure Index | Federal Reserve Bank of New York; method in Benigno et al. (2022), Staff Report 1017 | NY Fed website terms of use; the file is not redistributed here |
Models
| File | What it is | Where |
|---|---|---|
cerop_financial_xgb.pkl | Financial XGBoost, Week 5 configuration | models/ |
cerop_financial_preprocessor.pkl | StandardScaler for the 85 financial features | models/ |
cerop_operations_preprocessor.pkl | Week 4 operations ColumnTransformer (see the known mismatches) | models/ |
cerop_operations_rf.pkl | Tuned operations Random Forest, 1.24 GB | Available on request (see models/README.md) |
Code
| Folder | Contents |
|---|---|
notebooks/ | The ten notebooks above |
scripts/ | The original demo scripts (scoring, SHAP, labels, export checks) and the report-table extractor |
results/ | Week 5 and 6 result CSVs, both Week 6 runs, report tables, and the web exports |
console/ | The Risk Console (React + Vite) and the browser parity test |
site/ | This site, the deck, and the scripts that turn results into page data |
docs/ | Methodology, limitations, engineering log, notebook map |