当前位置:网站首页>蜥蜴书学习day1-机器学习概述
蜥蜴书学习day1-机器学习概述
2022-04-22 20:55:00 【Mail Lee】
1 概览
1.1 为什么使用机器学习
a)机器学习可以处理使用传统方法,太复杂,或者没有已知算法的问题
b)机器学习可以帮助人类学习,发现新趋势、新关联
1.2 机器学习应用例子
a)图像分类,用卷积神经网络CNN
b)自然语言处理NLP
c)强化学习(RL,reinforcement learn):给定环境(如游戏),训练代理(如机器人),选择在一段时间内将他们的奖励最大化(如,机器人会在玩家失去一些生命值时获得奖励),alphaGo就是用RL实现的

1.3 机器学习的分类
a)有监督和无监督学习:
有监督学习,训练集是有标记的,常见的有:
k-邻近算法
线性回归
逻辑回归
支持向量机
决策树和随机森林
神经网络
无监督学习:
聚类算法
k-均值算法
DBSCAN
分层聚类分析(HCA)
异常检测和新颖性检测
单类SVM
孤立森林
可视化和降维
主成分分析(PCA)
核成分分析
局部线性嵌入
t-分布随机近邻嵌入
关联规则学习
降维的目的是在不丢失太多信息的前提下,简化数据,方法是将多个相关特征合并为一个,也叫特征提取
b)批量学习和在线学习
批量学习:

在线学习(增量学习):
在线学习其实也是离线完成的,只是可以更新学习的数据流,在线学习的学习率越高,系统会迅速适应新数据,但同时会很快忘记就数据
c)基于实例的学习和基于模型的学习
基于实例的学习:系统通过实例的相似度来度量比较新实例和已经学习的实例,从而预测新实例的
基于模型的学习:构建实例的模型
1.4 机器学习的挑战
a)训练的数据量
b)训练数据不具有代表性
c)低质量数据
d)过拟合
解决过拟合的方法:
简化模型,减少参数
收集更多的数据
减少噪声
通过约束模型使其更简单,也被称为正则化:比如线性拟合中,规定截距为1
超参数:是学习算法的参数,不是模型的参数,在训练前确定,训练时不改变
e)欠拟合
1.5 测试和验证
分为测试集和训练集
防止调整超参数时,出现仅适用于测试集的最佳模型,却在实际应用中效果不佳,解决方法:保持验证,即选取测试集的一部分作为验证集,用剩下的测试集去训练具有多种超参数的模型,选择验证集表现最佳的超参数,再用完整的训练集去训练模型
版权声明
本文为[Mail Lee]所创,转载请带上原文链接,感谢
https://blog.csdn.net/qq_44849814/article/details/124230127
边栏推荐
- What are the types of documenter examination questions? How to prepare for the documenter examination of the construction department
- CmsEasy7. 6.3.2 logic vulnerabilities
- Leetcode-92-reverse linked list
- 对Swin-T中SW-MSA的一些理解
- Active mode and passive mode of FTP
- 2022 civil construction worker's question bank precision small question bank construction hall constructor
- 大量mapper IO优化(使用多线程异步+CountDownLatch)
- Semi synchronous replication of MySQL master-slave replication
- [200 opencv routines of youcans] 160 Otsu method of image processing
- MySQL开发技巧
猜你喜欢

Smart agriculture has become a development path, give full play to intelligence and liberate manpower

Minio基本使用与原理

Asynchronous replication of MySQL master-slave replication

UnityShader入门精要——素描效果渲染

Error running ‘JeecgSystemApplication‘: Command line is too long. Shorten command line for JeecgSyst

MySQL主从复制之GTID复制

基于SEIR模型的传染病预测软件开发

H. Maximal AND

MySQL advanced stored procedure storage function -- Introduction to stored procedure, basic syntax of stored procedure, variables (system variables, user-defined variables, local variables), if, param
Your so-called comfort is slowly destroying you!
随机推荐
动态数据库工具——Database Inspector
October's Android interview failed miserably in byte three, and fortunately won Xiaomi offer
MySQL 进阶 触发器 -- 触发器介绍、触发器语法、触发器案例
Mastering the tips of these references will help you get twice the result with half the effort~
Short link design and thinking
2020团队设计天梯赛(部分)
2020 team design ladder competition (part)
The list of SCI / SSCI journals has been updated and these journals have been eliminated~
Active mode and passive mode of FTP
buuctf-[Flask]SSTI
Virtual machine building and installation pulsar environment tutorial (for development and testing)
The interviewer would rather have my younger brother who has just graduated and worked for one year than me who has worked for five years, with an annual salary of 25W
What are the types of documenter examination questions? How to prepare for the documenter examination of the construction department
Asynchronous replication of MySQL master-slave replication
基于PAOGD_HW1的弹出的小球-简单建模、插值动画
[interview ordinary people vs Expert Series] please talk about the network quadruple
农村没网络怎样安监控,家里没有wifi安哪种监控器
String - 7 Output string (10 points) both printf function and puts function in C language standard function library can output string, but each has its own advantages and disadvantages. We combine the
Building a new generation of computing platform, stepvr will open the "door" of metauniverse in 2022
Ordinary functions as friends (using examples to solve friend functions)