监督学习与无监督学习
监督学习(Supervised Learning)
Ng的原文是:
We gave the algorithm a data set that the "right answers" were given.
即给定了一个正确结果的集合供算法学习,强调了需要实现准备好正负样本喂给机器。
无监督学习(Supervised Learning)
原文如下:
Given the data set, Unsupervised learning algorithm might decide that the data lives in two clusters, it breaks these data into two separated clusters, so it's called clustering algorithm.
可以对比发现,相比监督学习,无监督学习事先并未给出符合预期的正确结果,而是仅仅通过算法将这些数据进行聚类(clustering),但是我们事先并不知道归类的类别分别是什么。
分类(Classification)和回归(Regression)
原文:
The regression problems means a real, continuous value. However, the classification problems give the known several types and you need to classify. If the case is a small number of discrete values, we should treat it as a classification problem.
分类问题是用于将事物打上一个标签,通常结果为离散值[1]。例如区分一张图片中的动物是猫还是狗。而回归问题是需要预测具体值的,例如基于现有的房价数据,推测2020年的房价。其本质差别是待归属的类别是有限还是无限的(是否可穷尽)。
[1]:回归(regression)与分类(classification)的区别