High Cardinality Feature Encoding Questions

Compare methods to represent categorical features with large cardinality and reason about their stability and serving costs. Techniques include hashing, target encoding with out of fold schemes to prevent leakage, learned embeddings, frequency bucketing, and grouping rare categories. Discuss risks such as leakage and instability with daily arriving categories, strategies for unseen categories, computational and memory cost in serving, and how encoding choices interact with model type and retraining cadence.