3

目前我正在使用 glmnet 包来运行套索回归（在下面的示例中，它被保存到“fits”变量中。然后当我绘制 fit 变量时，它会正确出现，但系数标签非常小。任何想法如何我可以增加这些的大小吗？

下面的可重现示例...

require(glmnet)

#setup sample DF with 5 variables
set.seed(123)
sampleDF <- data.frame("V1"=rnorm(100,mean=0,sd=.10),"V2"=rnorm(100,mean=0,sd=.10),"V3"=rnorm(100,mean=0,sd=.10),"V4"=rnorm(100,mean=0,sd=.10),"V5"=rnorm(100,mean=0,sd=.10))

#break data into yVector & xMatrix to put into glmnet
yVector <- sampleDF[,1]
xMatrix <- as.matrix(sampleDF[,2:ncol(sampleDF)])

#use k-fold cross validation to find the min lambda
cv.glmmod <- cv.glmnet(xMatrix,yVector,alpha=1,nfolds=nrow(xMatrix),grouped=FALSE)
best_lambda <- cv.glmmod$lambda.min

#run glmnet
fits <- glmnet(xMatrix, yVector, family="gaussian", alpha=1, nlambda=100)

#plot results
plot(fits,label=TRUE,xvar="lambda")
4

1 回答 1

4

因为它看起来好像标签大小是硬编码的（并且全局更改cex将更改其他绘图功能），您可以更改plot.glmnet

# copy the plot function
myPlot <- plotCoef

# replace relevant part
body(myPlot)[[14]] <- quote(if (label) {
  nnz = length(which)
  xpos = max(index)
  pos = 4
  if (xvar == "lambda") {
    xpos = min(index)
    pos = 2
  }
  xpos = rep(xpos, nnz)
  ypos = beta[, ncol(beta)]
  text(xpos, ypos, paste(which), pos = pos, ...) # only changed this with ...
})

# copy first level of plot and replace plotCoef  with myPlot
newplotter <- plot.glmnet

body(newplotter)[[3]] <- quote(myPlot(x$beta, lambda = x$lambda, 
                                     df = x$df, dev = x$dev.ratio, 
                                        label = label, xvar = xvar, ...))

这应该增加文本（注意情节需要足够宽）

newplotter(fits,label=TRUE,xvar="lambda", cex=1.2)

在此处输入图像描述

于 2015-05-26T15:37:44.320 回答