机器学习(ML)策略 正交化(Orthogonalization)单一数字评估指标满足和优化指标(Satisficing and optimizing metrics)训练/开发/测试集划分(Train/dev/test distributions)开发集和测试集的大小(Size of dev and test sets)什么时候该改变开发/测试集和指标?(When to change de
文章目录 1. Why ML Strategy2. Orthogonalization3. Single number evaluation metric4. Satisficing and Optimizing metic5. Train/dev/test distributions6. Size of the dev and test sets7. When to change dev/