偏倚方差分解—

过拟合中模型复杂度和预测误差的曲线图：bias variance trade off

# bias variance trade off -------------------------------------------------set.seed(1)
test_size <- 50
sigma <- 0.5
test_x <- matrix(rnorm(test_size*35,0,1),test_size)
test_y <- ifelse(apply(test_x,1,sum)>0 , 1 , 0)library(glmnet)
pnum <- 35
train_error_all <- matrix(NaN,100,pnum)
test_error_all <- matrix(0,100,pnum)
for( i in (1:100)){flag = 1while (flag  == 1){train_x <- matrix(rnorm(50*35,0,1),nrow = 50)train_y <- ifelse(apply(train_x,1,sum)>0 , 1 , 0)lasso <- glmnet(train_x,train_y,alpha = 1, nlambda = 10000, family = 'gaussian',pmax = pnum)lambdas <- data.frame(df = lasso$df,lambda = lasso$lambda)ld <- aggregate(lambdas,by = list(lambdas$df),mean)ld <-  ld[-1,]if(dim(ld)[1] == pnum){flag = 0}}lasso1 <- glmnet(train_x,train_y,alpha = 1, lambda = ld$lambda, family = 'gaussian')result_train <- predict(lasso1, newx = train_x,type = 'response',s = ld$lambda)result_test <- predict(lasso1, newx = test_x,type = 'response',s = ld$lambda)train_error <- apply(abs(result_train - train_y),2,mean)test_error <- apply(abs(result_test - test_y),2,mean)train_error_all[i,ld$df] <- train_errortest_error_all[i,ld$df] <- test_errorprint(i)
}train_error_mean <- apply(train_error_all,2,mean)
test_error_mean <- apply(test_error_all,2,mean)library(ggplot2)
library(animation)
g1 <- ggplot()
saveGIF({for (i in 1:100){print(i)train_data <- data.frame(train_error_all[i,])train_data$num <- 1:35names(train_data) <- c('predict error','model complexity')train_data$type <- 'train_error'test_data <- data.frame(test_error_all[i,])test_data$num <- 1:35names(test_data) <- c('predict error','model complexity')test_data$type <- 'test_error'train_all <- data.frame(train_error)train_all$num <- 1:35names(train_all) <- c('predict error','model complexity')train_all$type <- 'average(train_error)'test_all <- data.frame(test_error)test_all$num <- 1:35names(test_all) <- c('predict error','model complexity')test_all$type <- 'average(test_error)'g1 <- g1  + geom_line(data = train_data,aes(x='model complexity',y='predict error'),lwd = 1,colour = 'lightblue') +geom_line(data = test_data,aes(x='model complexity',y='predict error'),lwd = 1,colour = 'lightpink') +geom_line(data = train_all,aes(x='model complexity',y='predict error'),lwd = 2,colour = 'blue') +geom_line(data = test_all,aes(x='model complexity',y='predict error'),lwd = 2,colour = 'red') print(g1)}
},movie.name='Bias-Variance-Trade-Off.gif',interval=0.5,ani.width=700,ani.height=600)######静态#########
for (i in 1:100){plot(1:pnum,train_error_all[i,],xlab = '',ylab = '',xlim = c(0,pnum),ylim = c(0,0.6),type = 'l',col = 'lightblue')par(new = T)plot(1:pnum,test_error_all[i,],xlab = '',ylab = '',xlim = c(0,pnum),ylim = c(0,0.6),type = 'l',col = 'lightpink')par(new = T)
}plot(ld$df,train_error,xlab = '',ylab = '',xlim = c(0,pnum),ylim = c(0,0.6),type = 'l',col = 'blue',lwd = 2)
par(new = T)
plot(ld$df,test_error,xlab = 'Model Complexity (df)',ylab = 'Prediction Error',xlim = c(0,pnum),ylim = c(0,0.6),type = 'l',col = 'red',lwd = 2)
box()

PS.本文代码是在一个公众号上看到的(名称我忘了T_T很久之前看的了)，当时做了笔记。后来找不到了，保存代码于此。

偏倚方差分解——R code相关推荐

R语言VAR模型方差分解代码
好的,您可以使用以下代码来进行 VAR 模型的方差分解: library(vars)# 建立 VAR 模型 model <- VAR(data, p = 2)# 进行方差分解 irf <- ...
R语言 VAR 模型的方差分解
可以使用以下代码来进行 VAR 模型的方差分解: library(vars)# 建立 VAR 模型 model <- VAR(data, p = 2)# 进行方差分解 irf <- irf ...
深度学习模型在训练集上很好而在测试集表现得不好而拟合次数并不多_机器学习中的过拟合，欠拟合和偏倚方差折衷...
过度拟合在机器学习中很重要. 很直观的解释过拟合:假设我们现在让机器学习考试做题,想象一种情况,机器逐字记住每个问题的答案(拟合非常好-完美).然后,我们可以在练习题上得分很高:我们这样做是基于希望实 ...
python 方差分解_干货：教你用Python来计算偏差-方差权衡
原标题:干货 :教你用Python来计算偏差-方差权衡作者:Jason Brownlee 翻译:吴振东本文约3800字,建议阅读8分钟. 本文为你讲解模型偏差.方差和偏差-方差权衡的定义及联系,并 ...
机器学习之过拟合与欠拟合以及偏差-方差分解
1.过拟合所谓过拟合就是:把训练样本自身的一些特点当作了所有潜在样本都会具有的一般性质,这样就会导致训练出的模型其泛化能力降低,这就是过拟合. 如何解决? 1)Early stopping Earl ...
UA MATH571A 回归分析概念与R code总结
UA MATH571A 回归分析概念与R code总结 Simple Linear Regression Multivariate Linear Regression Part 0 Basic R ...
var模型的matlab实现_Eviews中VAR模型的操作、脉冲响应分析和方差分解的实现
打开文件所在位置,获取数据.选中变量右键open打开var 操作EViews,在VAR对象的工具栏中选择"View"|"Lag Structure"|" ...
【机器学习－西瓜书】二、偏差-方差分解；泛化误差
2.5偏差与方差关键词:偏差-方差分解:泛化误差 . 偏差-方差分解是解释算法泛化性能的一种重要工具.偏差-方差分解试图对学习算法的期望泛化错误率进行拆解. 泛化误差可分解为:偏差,方差与噪声之和. ...
R code execution error处理
R code execution error 解决方法: Ctrl + Shift + F10 to restart your R session
偏差-方差分解，学习和验证曲线评估模型
偏差-方差分解参考链接:https://www.zhihu.com/question/20448464 https://blog.csdn.net/simple_the_best/article/d ...

偏倚方差分解——R code

过拟合中模型复杂度和预测误差的曲线图：bias variance trade off

偏倚方差分解——R code相关推荐

最新文章

热门文章