我已经使用SAS很长一段时间了，现在我想在R中翻译我的代码。我需要帮助来做以下事情：
为了更清晰，我编辑了这段代码。 我使用了很多for循环，我知道并不总是推荐。这个过程很慢
是否有代码/包（例如应用系列函数，“插入符号”包）可以使这个非常干净有效/快速，特别是当样本大小* bootsample＆gt; 1000万
非常感谢任何帮助。
samplesize <- 200
bootsize<- 500
myseed <- 123
#generating a fake dataset
id=1:n
set.seed(myseed)
x <- rnorm(samplesize, 5, 5)
y <- rnorm(samplesize, 2 + 0.4*x, 0.5)
data <- data.frame(id, x, y)
head(data)
id x y
1 1 2.197622 3.978454
2 2 3.849113 4.195852
3 3 12.793542 6.984844
4 4 5.352542 4.412614
5 5 5.646439 4.051405
6 6 13.575325 7.192007
# generate bootstrap samples
bootstrap <- function(nbootsamples, data, seed) {
bootdata <- data.frame() #to initialize it
set.seed(seed)
for (i in 1:nbootsamples) {
replicate <- i
bootstrapIndex <- sample(1:nrow(data), replace = TRUE)
datatemp <- data[bootstrapIndex, ]
tempall <- cbind(replicate, datatemp)
bootdata <- rbind(bootdata, tempall)
}
return(bootdata)
}
bootdata <- bootstrap(nbootsamples=bootsize, data=data, seed=myseed)
bootdata <- dplyr::arrange(bootdata, replicate, id)
head(bootdata)
#The data should look like this
replicate id x y
1 1 1 2.197622 3.978454
2 1 3 12.793542 6.984844
3 1 5 5.646439 4.051405
4 1 9 1.565736 3.451748
5 1 10 2.771690 3.081662
6 1 10 2.771690 3.081662
#Model-fitting and saving coefficient and means
modelFitting <- function(y, x, data) {
modeltemp <- glm(y ~ x,
data = data,
family = gaussian('identity'))
Inty <- coef(modeltemp)["(Intercept)"]
betaX <- coef(modeltemp)["x"]
sdy <- sd(residuals.glm(modeltemp))
data.frame(Inty, betaX, sdy, row.names = NULL)
}
saveParameters <- function(nbootsamples, data, seed) {
parameters <- data.frame() #to initialize it
for (i in 1:length(unique(data$replicate))) {
replicate <- i
datai <- data[ which(data$replicate==i),]
datatemp <- modelFitting(y, x,data=datai)
meandata <- data.frame(Pr_X=mean(datai$x))
tempall <- cbind(replicate, datatemp, meandata)
parameters <- rbind(parameters, tempall)
}
return(parameters)
}
parameters <- saveParameters(nbootsamples=bootsize, data=bootdata, seed=myseed)
head(parameters)
#Ultimately all I want is my final dataset to look like the following
replicate Inty betaX sdy Pr_X
1 1 2.135529 0.3851757 0.5162728 4.995836
2 2 1.957152 0.4094682 0.5071635 4.835884
3 3 2.044257 0.3989742 0.4734178 5.111185
4 4 2.093452 0.3861861 0.4921470 4.741299
5 5 2.017825 0.4037699 0.5240363 4.931793
6 6 2.026952 0.3979731 0.4898346 5.502320
答案 0 :(得分：2)
使用caret包可以轻松完成重采样回归。给出您的示例数据，通过广义线性模型运行200个引导程序样本的代码如下所示。
library(caret)
x = round(rnorm(200, 5, 5))
y= rnorm(200, 2 + 0.4*x, 0.5)
theData <- data.frame(id=1:200,x, y)
# configure caret training parameters to 200 bootstrap samples
fitControl <- trainControl(method = "boot",
number = 200)
fit <- train(y ~ x, method="glm",data=theData,
trControl = fitControl)
# print output object
fit
# print first 10 resamples
fit$resample[1:10,]
来自插入符号的输出如下所示：
> fit
Generalized Linear Model
200 samples
1 predictor
No pre-processing
Resampling: Bootstrapped (200 reps)
Summary of sample sizes: 200, 200, 200, 200, 200, 200, ...
Resampling results:
RMSE Rsquared MAE
0.4739306 0.9438834 0.3772199
> fit$resample[1:10,]
RMSE Rsquared MAE Resample
1 0.5069606 0.9520896 0.3872257 Resample001
2 0.4636029 0.9460214 0.3711900 Resample002
3 0.4446103 0.9549866 0.3435148 Resample003
4 0.4464119 0.9443726 0.3636947 Resample004
5 0.5193685 0.9191259 0.4010104 Resample005
6 0.4995917 0.9451417 0.4044659 Resample006
7 0.4347831 0.9494606 0.3383224 Resample007
8 0.4725041 0.9483434 0.3716319 Resample008
9 0.5295650 0.9458453 0.4241543 Resample009
10 0.4796985 0.9514595 0.3927207 Resample010
>
caret GitHub site提供了有关如何使用插入符号的详细信息，包括生成的模型对象的内容（例如，访问各个模型，以便您可以使用predict()函数生成用于模拟的预测）。
Caret还支持并行处理。有关如何使用插入符号并行处理的示例，请阅读Improving Performance of Random Forest with caret::train()。
此外，R中通过R中的Monte Carlo包支持蒙特卡罗模拟。