Loop - LASSO 从模拟数据中提取系数

Question

我试图从模拟数据的 LASSO 实验中找出哪些系数被正确和不正确地收缩为 0。我可以让它在一次迭代中工作，但我不知道如何正确地循环它以便我可以提取 100 次迭代的相关信息。这是我当前的代码

library(MASS)
library(glmnet)
N=100
n=200
p=200


set.seed(123)
f.non<-data.frame(NULL)
f.disc<-data.frame(NULL)

  X= mvrnorm(200, rep(0,p), diag(1,p,p))
  Y=rowSums(X[,1:10])+3.5*rnorm(n)
  lasso.model<-cv.glmnet(X,Y,alpha=1,intercept=FALSE)
  lasso.coef<-coef(lasso.model,s=lasso.model$lambda.1se)
  f.non<- sum(lasso.coef[2:11,] == 0)
  f.disc <- sum(lasso.coef[12:201,] != 0)

}

上面的部分是我尝试通过设置一个空数据框进行循环，然后在循环中捕获结果但我无法让循环工作

Answer 1

因此，如果我理解正确（但我不确定我是否理解正确），您正在尝试提取 lasso.coef 对象的值（和名称？），具体取决于它们是否重新等于零。

你可以像这样在 for 循环中重复 100 次（如果你想收集变量的名称）：

f.non <- c(NULL)
f.disc <- c(NULL)
N=100
for (i in 1:N) {
  n=200
  p=200
  X = mvrnorm(200, rep(0, p), diag(1, p, p))
  Y = rowSums(X[, 1:10]) + 3.5 * rnorm(n)
  lasso.model <- cv.glmnet(X, Y, alpha = 1, intercept = FALSE)
  lasso.coef <- coef(lasso.model, s = lasso.model$lambda.1se)
  # lasso_coefs <- lasso.coef[2:201]
  # setNames(lasso_coefs, paste("V", as.character(seq(2, 201, 1)), sep = ""))
  lasso_coefs_df <- data.frame(coef = lasso.coef[2:201], name = paste("V", as.character(seq(2,201,1)), sep = ""))

  # f.non[[i]] <- as.list(lasso_coefs[lasso_coefs == 0])
  # f.disc[[i]] <- as.list(lasso_coefs[lasso_coefs != 0])
  f.non[[i]] <- lasso_coefs_df[lasso_coefs_df$coef == 0, ]
  f.disc[[i]] <- lasso_coefs_df[lasso_coefs_df$coef != 0, ]
}

否则，更类似于您的尝试，如果您只想要每个变量范围 (2-11、12-201) 的总和：

f.non <- c(NULL)
f.disc <- c(NULL)
N=100
for (i in 1:N) {
  n=200
  p=200
  X = mvrnorm(200, rep(0, p), diag(1, p, p))
  Y = rowSums(X[, 1:10]) + 3.5 * rnorm(n)
  lasso.model <- cv.glmnet(X, Y, alpha = 1, intercept = FALSE)
  lasso.coef <- coef(lasso.model, s = lasso.model$lambda.1se)
  f.non[[i]] <- sum(lasso.coef[2:11,] == 0)
  f.disc[[i]] <- sum(lasso.coef[12:201,] != 0)
}

Loop - LASSO 从模拟数据中提取系数

Loop - LASSO Coefficient extraction from simulation data

r

dataframe

for-loop

coefficients

lasso-regression