Skip to content

week1

Jump to bottom
zhuangyongyao edited this page Aug 16, 2016 · 2 revisions

机器学习斯坦福公开课学习笔记

from 2016.05.09

link: https://www.coursera.org/learn/machine-learning

resource: https://share.coursera.org/wiki/index.php/ML:Main

其他有用资源: https://share.coursera.org/wiki/index.php/ML:Useful_Resources

testcase: https://www.coursera.org/learn/machine-learning/discussions/all/threads/0SxufTSrEeWPACIACw4G5w

1. 第一周

Tom Mitchell provides a more modern definition: "A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P, improves with experience E."

Example: playing checkers.

  • E = the experience of playing many games of checkers
  • T = the task of playing checkers.
  • P = the probability that the program will win the next game.

入门,什么是机器学习:

  • 垃圾邮件处理
  • 自然语言处理
  • 搜索引擎

1.1 supervised learning 监督学习

  • 回归问题. 房屋价格评估, 给定了房屋面积和价格,预测给定的面积对应的价格.

  • 分类问题. 肿瘤推测.

  • 垃圾邮件分类.

  • 根据病人检查结果是否有糖尿病的关系的历史分类, 来判断新病人是否有糖尿病.

Training set.

训练集, 根据训练集

线性回归!!!

1.2 无监督学习

  • 聚类算法
  • social network analysis
  • 网页分类
  • market segmentation 市场划分
  • orgnize computing clusters
  • 天文数据分析 Astonomical data analysis

cocktail party problem 鸡尾酒问题算法,分离声音

[w,s,v] = svn((repment(sum(x.*x, 1), size(x, 1), 1).*x) *x');

svn 是什么? 奇异解,解线性方程, 在 octave 里面有直接的函数

工具

octave

在工程上一般都先用octave 进行模拟,建模,这样速度会非常快,会更有效率学习一些机器学习的东西. 等需要实用的时候,再使用c++或者java进行实现.

model and cost function

training set -> learning algorithm 
                    |
                    v
size of house->     h   ->  estimated price   
       
h is the hypothesis. h is the maps from x's to y's

h(x) = ø0 + ø1x

hypothosis: 假设

cost function

线性回归. 回归问题,通过训练集,来确定.

代价函数,平方差代价函数,求平方误差代价函数的最小值.

代价函数是干什么的

Clone this wiki locally