8.2 Locally weighted linear regression(LWLR)
对邻近的数据点加大权重,对远点减小权重。LWLR回归的预测曲线如下图:
顶部的图有些underfitting,底部的图有些overfittting。
下一个section会讲如何定量地(quantitatively)描述underfitting和overfittting
扫描二维码关注公众号,回复:
2623822 查看本文章
8.4 Shrinking coefficients to understand our data
引入原因是:when we have more features than data points, then we cannot matrix inverse to retrieve the regression results.
解决办法:总的来说是shrinkage methods. 而shrinkage methods包括两类:一是ridge regression或者lasso;二是forward stagewise regression,算法给出的结果也和lasso相近。
除了完成引入原因的问题外,shrinking还可以用于throw out some unimportant features
8.5 Bias-Variance tradeoff
中文翻译是偏差-方差,有篇文章解释的比较好:http://blog.csdn.net/qq_30490125/article/details/52401773
8.6.2
实例中有如何完成cross-validation的实现