kaggle-feature-engineering

Automate groupby aggregations and target encoding for Kaggle tabular data by running /kaggle-feature-engineering to generate group-based, interaction, and CV-safe categorical features.

Yes, you can use GPU acceleration for feature engineering with pandas and cuDF. The Skill utilizes pandas, cupy, and cuDF dependencies to accelerate feature generation for large datasets.

Generate time series lag and rolling features for machine learning competitions by executing the Skill, which builds temporal features specifically designed to improve model cross-validation performance.

You do not strictly need a GPU to run kaggle-feature-engineering for interaction feature generation, but having GPU support via cupy and cuDF significantly accelerates the feature creation process.

CV-safe target encoding for categorical features in Kaggle competitions works by transforming categories using out-of-fold target means, preventing data leakage and improving model generalization.

Feature selection matters after batch feature generation in model optimization because the Skill facilitates selecting relevant features post-batch, reducing dimensionality and enhancing model performance.