Python机器学习手册：从数据预处理到深度学习

16.4

　在超大数据集上训练分类器

｜

267

#

训练模型

model = logistic_regression.fit(features_standardized, target)

讨论

正则化（

regularization

）是一种通过惩罚复杂模型来减小其方差的方法。准确地说，就

是将一个罚项加在我们希望最小化的损失函数上，通常称为

L1

惩罚和

L2

惩罚。

L1

惩

罚是 ：

α

∑

j =1

p

β

j

where β

j

is the parameters of the jth of p features being learned and α is a hyperpara‐

meter denoting the regularization strength. With the L2 penalty:

α

∑

j =1

p

β

j

2

Higher values of α increase the penalty for larger parameter values (i.e., more com‐

plex models). scikit-learn follows the common method of using C instead of α where

C is the inverse of the regularization strength:

C =

1

α

. To reduce variance while using

logistic regression, we can treat C as a hyperparameter ...