R中的有效循环逻辑回归 [英] efficient looping logistic regression in R

问题描述

我正在尝试对约40万个预测变量进行多个logistic回归分析.我想将每次运行的输出捕获到输出表的行/列中.

I'm trying to run multiple logistic regression analyses for each of ~400k predictor variables. I would like to capture the outputs of each run into a row/column of an output table.

我的数据分为两个部分.我有一个400000 x 189双矩阵(mydatamatrix)，其中包含在189个个体(P1)中测得的我的400000个预测变量的观察值/数据.我还有第二个189 x 20数据帧(mydataframe)，其中包含结果变量和另一个预测变量(O1和P2)以及此特定分析中未使用的其他18个变量.

My data organised in two parts. I have a 400000 x 189 double matrix (mydatamatrix) that contains the observations/data for each of my 400000 predictor variables measured in 189 individuals (P1). I also have a second 189 x 20 data frame (mydataframe) containing the outcome variable and another predictor variable (O1 and P2) plus 18 other variables not used in this particular analysis.

我的回归模型是O1~ P1+P2，其中O1是二进制.

My regression model is O1~ P1+P2, where O1 is binary.

我有以下工作循环:

为结果创建输出文件

output<-data.frame(matrix(nrow=400000, ncol=4))
names(output)=c("Estimate", " Std. Error", " z value", " Pr(>|z|)")

运行i预测变量的逻辑回归循环并将输出存储在输出文件中

run logistic regression loop for i predictors and store output in output file

for (i in c(1:400000)){
  result<-(glm(mydataframe$O1 ~ mydatamatrix[,i] + as.factor(mydataframe$P2),
               family=binomial))
  row.names(output)<-row.names(mydatamatrix)
  output[i,1]<-coef(summary(result))[2,1]
  output[i,2]<-coef(summary(result))[2,2]
  output[i,3]<-coef(summary(result))[2,3]
  output[i,4]<-coef(summary(result))[2,4]
}

但是，运行时间非常长(输出前20k个测试花了一个多小时).有没有更有效的方法来进行此分析?

However, the run time is huge (it took over an hour to output the first 20k tests). Is there a more efficient way to run this analysis?

推荐答案

如果使用apply而不是for循环，它将更快:

It will be faster if you use apply instead of a for loop:

t(apply(mydatamatrix, 2,
        function(x)
          coef(summary(glm(mydataframe$O1 ~ x + as.factor(mydataframe$P2), 
                           family=binomial)))[2, 1:4]))

这篇关于R中的有效循环逻辑回归的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！