知识卡片
持续 ML 系统与一次性训练的分野
前面的讨论有时把模型当作「训练然后部署、一次且仅一次」的东西,更深一层的看法是区分「训练一次即部署」的模型与「以更连续方式训练」的模型,后者称为持续 ML 系统。典型持续 ML 系统以流式或周期批次方式接收新数据,用它触发训练出一个更新后的模型版本,再把该版本推送到服务。作者强调:从 MLOps 视角看,一次性训练的模型与持续更新的模型之间存在重大差异——转向持续 ML 会抬高自动化验证的门槛,引入反馈回路与模型对外部世界变化的反应这类麻烦,管理连续数据流、应对模型失效与损坏,乃至「给模型引入新特征来训练」这种看似琐碎的事,全都会增加系统复杂度。理解这个分野是本章全部后续讨论的前提:不是把同一套流程跑得更频繁,而是一类系统性质不同的工程对象。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
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
- Continuous ML · 查看原书依据
Typical continuous ML systems receive new data in a streaming or periodic batch fashion and use this to trigger training an updated model version that is pushed to serving. Clearly, there are major differences from an MLOps perspective between a model that is trained once versus a model that is updated in a continuous manner.
参考来源
- 位置:《Reliable Machine Learning》第10章《Continuous ML》(源文件:extracted/Reliable Machine Learning/_epub-src/OEBPS/ch10.xhtml)
- 结论依据:原文明确阐述该主题,并给出定义、实践或反例。
- 原始内容:Typical continuous ML systems receive new data in a streaming or periodic batch fashion and use this to trigger training an updated model version that is pushed to serving. Clearly, there are major differences from an MLOps perspective between a model that is trained once versus a model that is updated in a continuous manner.