What problem does it solve? Building effective features for tabular machine learning benchmarks requires domain knowledge, leakage-safe encoding patterns, and iterative evaluation, which is slow and error-prone when done manually. ## Core Features & Use Cases - Dataset-Specific Knowledge: Provides tailored feature engineering directions for the Taobao ad conversion dataset (Bayesian-smoothed CVR, interaction CVR, log-count frequencies) and the Dia diabetes dataset (BMI thresholds, composite risk scores, cross-validated target encoding). - Leakage-Safe Code Patterns: Supplies reusable Python snippets for KFold target encoding, Bayesian smoothing, and history-log aggregation that prevent target leakage. - Iterative Evaluation Loop: Guides AUC-driven feature selection with early stopping via the benchmark's evaluate API. - Use Case: Given a Taobao CTR prediction task, use this Skill to build entity CVR features from the history log, evaluate them against the benchmark, and ablate toward a lean high-AUC feature set. ## Quick Start Use this skill to iteratively engineer and evaluate features that maximize AUC on the provided tabular benchmark dataset.
