引导式测验 · 中等 等级
功能商店测验
练习特征存储历史记录、在线新鲜度和实体连接,包括共享定义如何在不保证相同输入的情况下减少偏差。
问题 1 的 8
Why use shared feature definitions for training and online serving?
预览每个问题8 问题
当您参加测验时,就会出现答案和解释。
- Why use shared feature definitions for training and online serving?
- To guarantee every cached value is always fresh
- To reduce skew caused by separate transformation implementations
- To eliminate the need for timestamps
- To choose a GPU model automatically
- 线下商店和网上商店有什么区别?
- 离线保存历史数据用于训练;在线快速提供最新值以进行实时预测
- 离线用于图像,在线用于文本
- 它们是相同的但名称不同
- 周末线下运行,工作日线上运行
- 为什么时间点连接在构建训练数据时很重要?
- 它们使数据库运行得更快
- 他们将功能翻译成其他语言
- 他们通过使用每个事件发生时的特征值来防止数据泄漏
- 它们减少了模型的文件大小
- Which storage pattern commonly supports low-latency feature lookup by entity key?
- An archival tape read for every request
- A printed document catalog
- A model-weight-only file
- A key-value online store such as Redis or DynamoDB
- 在这种情况下,什么是“特征”?
- 来自数据的输入变量,例如 30 天内的平均支出
- 模型代码中的错误
- 仪表板的配色方案
- GPU 的一种
- 共享功能目录可以提供哪些额外好处?
- 它每小时自动删除旧数据
- 它取代了任何模型训练的需要
- 它可以让为一个模型构建的功能被其他模型发现和重用
- 它加密公司的电子邮件
- A feature’s event time precedes a loan decision, but the value only arrived afterward. What must a historical training join consider?
- Only whether its column name is short
- A rule that every past event was already known
- Only the current online value
- Whether that value was actually available at prediction time
- An online feature has not been refreshed after new events arrived. Which property should a feature-store monitor inspect?
- The font size of the registry page
- Feature freshness relative to the prediction’s needs
- The model’s number of attention heads only
- Whether the feature name contains a digit