知识卡片
训练管线静默失败
训练管线具备普通数据管道的全部可靠性挑战(缺数据、格式错、软件 bug、配置错、资源不足、硬件/分布式失败),还有 ML 特有风险:因数据分布漂移、数据缺失、欠采样或无法使用部分子集而“静默失败”——输出看似正常但已偏离。这正是它比普通 ETL 更难可靠运行的原因,相关监测见第 7、9 章。
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
- · 训练管线
原书出处 · 1 处
- Introduction · 查看原书依据
But ML models can fail silently for reasons related to data distribution, missing data, undersampling, or a whole host of problems unknown in the regular ETL world.3 One concrete example, covered in more detail in Chapter 2, hinges on the idea that missing, misprocessing, or otherwise not being able to use subsets of data is a common cause of failure for ML training pipelines. We’ll talk about ways to monitor training pipeline
参考来源
- 位置:《Reliable Machine Learning》第01章《Introduction》(源文件:extracted/Reliable Machine Learning/_epub-src/OEBPS/ch01.xhtml)
- 结论依据:原文列举普通 ETL 失败模式后,强调 ML 会因数据分布等问题静默失败,是更难运维的根因。
- 原始内容:But ML models can fail silently for reasons related to data distribution, missing data, undersampling, or a whole host of problems unknown in the regular ETL world.3 One concrete example, covered in more detail in Chapter 2, hinges on the idea that missing, misprocessing, or otherwise not being able to use subsets of data is a common cause of failure for ML training pipelines. We’ll talk about ways to monitor training pipeline