car_lemons

Case study comparing prediction methods with re-usable template code

Example research question: predict if car purchased at an auction is a kick to provide best inventory selection to their customers from Kaggle

Models

• Random Forest tuned using random search
  • Intuition: RF is a good choice for data with a high number of observations vs. features. It is a low bias, high variance method. It has good interpretability vs. SVM, and potentially better performing than a Lasso model due to its nonlinear nature.
• KNN
• Kernel SVM
• Light Gradient Boosted Machine
• Lasso

Development

Data

In the data/ folder, run:

kaggle competitions download -c DontGetKicked

or download manually.

Environment

• pipenv

Testing

• pytest

Contributing

CI: formatting, pytests pass & coverage, HISTORY updated