当前位置:网站首页>机器学习之评价指标(一)——回归评价指标
机器学习之评价指标(一)——回归评价指标
2022-07-27 15:49:00 【helpburn】
机器学习可以分为回归、分类、聚类、降维等。不同任务有自己的评价指标,下面我们先介绍一下回归的评价指标。
回归问题
如果预测变量是连续的我们称为回归,回归中如果只有一个自变量和一个因变量,二者关系可以用一条直线近似表示,这种回归称为一元线性回归;如果有两个及两个以上的自变量,艾尔自变量与因变量是线性关系则称为多元线性回归。回归属于监督学习。
回归(也叫拟合)问题比较简单,衡量指标也相对好理解。我们用yi表示真实值, y ^ \widehat{y} yi表示预测值。
平均绝对误差
平均绝对误差MAE(Mean Absolute Error),又称为L1范数损失(L1-norm loss)。取真实值与预测值差的绝对值的和,然后求平均。
MAE(y, y ^ \widehat{y} y) = 1 n ∑ i = 1 n \frac{1}{n} \sum_{i=1}^{n} n1∑i=1n |yi - , y ^ \widehat{y} yi|
均方误差
平均平方误差MSE(Mean Squared Error),又称为L2范数损失(L2-norm loss)。取真实值与预测值差的平方和,然后求平均。
MSE(y, y ^ \widehat{y} y) = 1 n ∑ i = 1 n \frac{1}{n} \sum_{i=1}^{n} n1∑i=1n (yi - , y ^ \widehat{y} yi)2
均方根误差(标准误差)
均方根误差RMSE(Root Mean Squarde Error),RMSE对一组数据中特别大或者特别晓得误差特别敏感,因此该指标能够很好的反映出误差的精密度,这也就使其鲁棒性较差。
公式的根号打不出来啊!!!
平均绝对百分比误差
平均绝对百分比误差(MAPE, Mean Absolute Percentage Error):是对绝对误差和真值百分比的误差损失预期。

决定系数 R Squared
决定系数(coefficient of determination)也叫拟合优度或者R2分数,反应自变量x对因变量y的变动的解释程度,越接近1,模型拟合的越好。例如,如果决定系数是0.86,这就意味着y的86%的变化可以通过x解释,可以解释其他的变化没有包含到模型中。



边栏推荐
- 快解析结合海典医药
- 交换机和路由器技术-03-交换机基本配置
- The chess robot broke the finger of a 7-year-old boy. Netizen: it violated the first law of robots
- An analysis of CPU explosion of a smart logistics WCS system in.Net
- Why is domestic Xinguan oral medicine a drug for the treatment of AIDS
- How to learn C language? This article gives you the complete answer
- Hegong sky team vision training Day8 - vision, target recognition
- Establishing SSL connection without server‘s identity verification is not recommended
- VO、DO、DTO、PO是什么
- Redis: configuring AOF does not work
猜你喜欢

Cow! His secret is to reproduce the paper in 2 hours——

Because the employee set the password to "123456", amd stolen 450gb data?

Can deep learning overturn video codec? The first prize winner of the National Technological Invention Award nags you in the little red book

DDD (Domain Driven Design) layered architecture

诸神黄昏,“猫抖快”告别大V时代

Initial polymorphism

Chery omenda is also too similar to Chang'an uni-t, but does it look like it? Is the product power like it?

Kubernetes Part 7: using kubernetes to deploy prometheus+grafana monitoring system (kubernetes work practice class)

20 years ago, he was Ma Yun's biggest enemy

Mlx90640 infrared thermal imager temperature sensor module development notes (VII)
随机推荐
Mysql database defines cursor in trigger
选择体育场馆的LED显示屏时应该注重哪些方面
面试官:什么是脚手架?为什么需要脚手架?常用的脚手架有哪些?
微信小程序 云函数批量删除多条数据 Error: errCode: -502005 database collection not exists
Taishan Office Technology Lecture: word strange paragraph borders
微信小程序 实现位置地图显示,引入map地图,不含导航
MySQL: Functions
科目三: 直线行驶
一文理解分布式开发中的服务治理
20 years ago, he was Ma Yun's biggest enemy
ES6 array method and pseudo array to array method
How to learn C language? This article gives you the complete answer
【单片机】2.1 AT89S52单片机的硬件组成
Wechat applet cloud function batch delete multiple data error: errcode: -502005 database collection not exists
微信小程序 实现拨打电话
微软默默给 curl 捐赠一万美元,半年后才通知
【单片机】2.2 AT89S52的引脚功能
What are VO, do, dto and Po
详解二叉树之堆
Database hyperphone (III)