知识卡片
训练系统的必备要素
一个可用训练系统至少需要五类东西,顺序可变、也可合并:训练数据(含人工标注,应已预处理标准化、以适合训练高效访问的格式存放,并置于访问受控且执行策略的环境);模型配置系统(把单模型配置与系统整体配置分开);模型训练框架(编排、作业/工作调度、训练或模型开发软件);模型质量评估系统;以及把可服务模型同步到下一阶段(通常是服务系统)的机制。具备这些才提供最低限度可用的技术环境;在此之上还应为可靠性与可管理性补上多级流水线监控、特征与模型的团队归属元数据、完整特征存储系统。注意「高效访问」依模型而不同,不是统一格式要求。
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
- Training Systems · 查看原书依据
Requirements A training system requires the following elements, although they might appear in a different order or combined with one another: Data to train on This includes human labels and annotations if we have them. This data should be preprocessed and standardized by the time we use it.
参考来源
- 位置:《Reliable Machine Learning》第07章《Training Systems》(源文件:extracted/Reliable Machine Learning/_epub-src/OEBPS/ch07.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:Requirements A training system requires the following elements, although they might appear in a different order or combined with one another: Data to train on This includes human labels and annotations if we have them. This data should be preprocessed and standardized by the time we use it.