知识卡片

普通读书笔记卡 · 7004.g · 生成模型:DeepSeek-V4.1-Flash

数据摄取

理解等级:初级

摄取(Ingestion)是把数据接收进系统并写入存储以便后续处理,此阶段必然发生过滤与选择——并非所有生成的数据都会被采集;当数据过多时还可采样以省成本,但须衡量采样的质量代价并按时间/分片比例采样,避免突发期丢细节。最佳实践至少提供简单 API 来摄取,以便确认收讫、记录日志并施加治理策略。可靠性重点在正确性与吞吐:误放数据真实发生(如 off-by-one 导致日期目录错位)。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • Data Management Principles · 查看原书依据

      In most cases, we will want to provide data ingestion via at least a simple API because this provides an obvious place to acknowledge receipt/storage of the data, log the ingestion, and apply any governance policies about the data. Reliability concerns during the ingestion phase typically focus on correctness and throughput.

    参考来源

    - 位置:《Reliable Machine Learning》第02章《Data Management Principles》(源文件:extracted/Reliable Machine Learning/_epub-src/OEBPS/ch02.xhtml) - 结论依据:原文说明摄取即过滤/采样,并强调 API 化与正确性是关键可靠性关注点。 - 原始内容:In most cases, we will want to provide data ingestion via at least a simple API because this provides an obvious place to acknowledge receipt/storage of the data, log the ingestion, and apply any governance policies about the data. Reliability concerns during the ingestion phase typically focus on correctness and throughput.