知识卡片
欠拟合也可能来自训练与表示限制
训练及独立数据都表现差,可能因为容量、特征或训练不足,也可能有噪声和任务问题。正则、早停及降维都有取舍,过强会丢信号;不能把任何调强正则都当泛化改善。
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
- · 查看原书依据
You can use regularization, early stopping, and dimension reduction to reduce overfitting and improve accuracy on new examples. But if you increase the regularization strength or reduce the dimensions of your model too much, you may overgeneralize and underfit, as in figure D.3.
参考来源
- 位置:《Natural Language Processing in Action, Second Edition》附录D《附录D:机器学习工具与技术》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/appendix-d.html)
- 结论依据:摘录中关于欠拟合也可能来自训练与表示限制的定义、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:You can use regularization, early stopping, and dimension reduction to reduce overfitting and improve accuracy on new examples. But if you increase the regularization strength or reduce the dimensions of your model too much, you may overgeneralize and underfit, as in figure D.3.