知识卡片
调参表不能把测试集变成训练反馈
保存每个表示、模型与指标,便于比较流水线并发现训练与验证差距。多次依同一集合选主题数或算法,会把该集合用作验证集;应另保留最终测试集,才能评价选型后的整体效果。
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
- · 从词数到语义分析
原书出处 · 1 处
- · 查看原书依据
Unless you have a perfect memory, by now, you must be pretty annoyed by scrolling or paging back to the performance of the previous model. And when you’re doing real-life NLP, you’ll probably be trying many more models than in our toy example. That’s why data scientists record their model parameters and performance in a hyperparameter table.
参考来源
- 位置:《Natural Language Processing in Action, Second Edition》第04章《从词数到语义分析》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/chapter-4.html)
- 结论依据:摘录中关于调参表不能把测试集变成训练反馈的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。
- 原始内容:Unless you have a perfect memory, by now, you must be pretty annoyed by scrolling or paging back to the performance of the previous model. And when you’re doing real-life NLP, you’ll probably be trying many more models than in our toy example. That’s why data scientists record their model parameters and performance in a hyperparameter table.