build-feature-store

Implement centralized feature management with Feast for consistent feature serving across training and inference.

Expected: Feast repository initialized with config file, sample feature definitions created, offline and online stores configured, registry path accessible.

On failure: Verify database/Redis credentials (psql -U feastuser -h localhost), check connection strings format, ensure databases exist (CREATE DATABASE featurestore), verify cloud permissions for S3/BigQuery/DynamoDB, test connectivity to storage backends, check Feast version compatibility with backends (feast version).

使用 Feast 建置特徵儲存以進行集中特徵管理，配置離線和線上儲存以進行批次和即時服務，透過轉換定義特徵視圖，並為 ML 管道實現時間點正確連接。在管理多個 ML 模型的特徵、確保訓練服務一致性、為即時推理提供低延遲特徵、跨專案重用特徵定義或建置用於發現和治理的特徵目錄時使用。 來源：pjt222/development-guides。