Give Me Some Credit

Credit default risk prediction pipeline — Kaggle competition dataset.

Architecture

Notebook (.py)  →  presentation layer (Jupytext)
Python package  →  logic (data, features, models)
Papermill       →  parameterized execution & automation

Project layout

credit/   # installable package — all logic lives here
  data/                #   load_train(), load_test()
  features/            #   build_features()
  models/              #   train(), evaluate()
  utils/               #   get_logger()
notebooks/             # Jupytext percent-format .py files (version-control friendly)
  01_eda.py
  02_feature_engineering.py
  03_model_training.py
  04_evaluation.py
outputs/               # Papermill-executed .ipynb outputs (gitignored except .gitkeep)
data/                  # raw CSVs (gitignored)

Setup (macOS / Linux)

poetry install

Activate the virtual environment (Poetry 2.0+):

# Option 1 — recommended
source $(poetry env info --path)/bin/activate

# Option 2 — one-off commands without activating
poetry run python ...

Install the Jupyter kernel for this env:

python -m ipykernel install --user --name give-me-some-credit

Setup (Windows)

Install Poetry via PowerShell:

(Invoke-WebRequest -Uri https://install.python-poetry.org -UseBasicParsing).Content | python -

If PowerShell blocks script execution, allow it first:

Set-ExecutionPolicy -ExecutionPolicy RemoteSigned -Scope CurrentUser

Then install dependencies and activate the environment:

poetry install

# activate
& $(poetry env info --path)\Scripts\Activate.ps1

# or run one-off commands without activating
poetry run python ...

Install the Jupyter kernel:

python -m ipykernel install --user --name give-me-some-credit

Interactive development (JupyterLab)

jupytext ships with a JupyterLab extension — open .py notebooks directly as notebooks, write code, and run cells without any conversion step:

jupyter lab notebooks/

Right-click a .py file → Open With → Notebook. Edits are saved back to the .py file automatically.

Automated / parameterized runs (Papermill)

Use Papermill for batch execution with overridden parameters (e.g. in CI or experiments). It requires a .ipynb input, so convert first:

jupytext --to notebook notebooks/03_model_training.py

papermill notebooks/03_model_training.ipynb outputs/03_model_training.ipynb \
  -p n_estimators 200 \
  -p max_depth 5

Data

Place the Kaggle files in data/:

• cs-training.csv
• cs-test.csv

Download from: https://www.kaggle.com/competitions/GiveMeSomeCredit/data