知识卡片
训练拟合优度不代替泛化评价
预测与观察越接近可表示更好拟合,但如果模型同时学到样本偶然波动,新数据误差仍可能很大。先说明拟合针对训练还是独立评价数据,再比较误差,不把一张贴合曲线当部署证明。
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
- · 机器学习基础
原书出处 · 1 处
- · 查看原书依据
The goodness-of-fit refers to how closely a model’s prediction matches the observed (true) values. Overfitting, Underfitting, and Bias-Variance Trade-Off Underfitting refers to the case when a model is too simple or has a small number of features or regularized too much. These make the model inflexible in learning from the data set. Underfitting occurs when the model fails to capture important patterns in the training
参考来源
- 位置:《Machine Learning Fundamentals: Concepts, Models, and Applications》第02章《机器学习基础》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Machine Learning Fundamentals: Concepts, Models, and Applications/_epub-src/OEBPS/xhtml/13_chapter02.xhtml)
- 结论依据:摘录中关于训练拟合优度不代替泛化评价的定义、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:The goodness-of-fit refers to how closely a model’s prediction matches the observed (true) values. Overfitting, Underfitting, and Bias-Variance Trade-Off Underfitting refers to the case when a model is too simple or has a small number of features or regularized too much. These make the model inflexible in learning from the data set. Underfitting occurs when the model fails to capture important patterns in the training