machine learning之logistic regression

machine learning

目录

[TOC]

第三周—logistic regression

解决overfitting的问题—regularization

Small values for parameters will lead to simpler hypothesis and avoid overfitting.
在cost function中加入额外的penalty，即regularization term:

the regularization parameter $\lambda$ is important. Too high value for $\lambda$ may result in underfitting.

接下来就是在linear regression和logistic regression中运用regularization.

regularized linear regression

regularized logistic regression

最初的cost function：

第四周—neural network

为什么需要了解neural network呢？现实场景中的分类问题通常会有很多变量，导致feature的数目巨增，一般平方feature数目为$\frac{n^2}{2}$，n为变量数，这时利用logistic regression就会十分复杂和耗时。neural network能够很好地解决这种问题。