broom (version 0.3.4)

cv.glmnet_tidiers: Tidiers for glmnet cross-validation objects

Description

Tidying methods for cross-validation performed by glmnet.cv, summarizing the mean-squared-error across choices of the penalty parameter lambda.

Usage

## S3 method for class 'cv.glmnet':
tidy(x, ...)
## S3 method for class 'cv.glmnet':
glance(x, ...)

Arguments

x

a "cv.glmnet" object

...

extra arguments (not used)

Value

• All tidying methods return a data.frame without rownames, whose structure depends on the method chosen.

  tidy produces a data.frame with one row per choice of lambda, with columns

• lambda
  penalty parameter lambda
• estimate
  estimate (median) of mean-squared error or other criterion
• std.error
  standard error of criterion
• conf.high
  high end of confidence interval on criterion
• conf.low
  low end of confidence interval on criterion
• nzero
  number of parameters that are zero at this choice of lambda
• glance returns a one-row data.frame with the values
• nulldev
  null deviance
• npasses
  total passes over the data across all lambda values

Details

No augment method exists for this class.

Examples

if (require("glmnet", quietly = TRUE)) {
    set.seed(2014)

    nobs <- 100
    nvar <- 50
    real <- 5

    x <- matrix(rnorm(nobs * nvar), nobs, nvar)
    beta <- c(rnorm(real, 0, 1), rep(0, nvar - real))
    y <- c(t(beta) %*% t(x)) + rnorm(nvar, sd = 3)

    cvfit1 <- cv.glmnet(x,y)

    head(tidy(cvfit1))
    glance(cvfit1)

    library(ggplot2)
    tidied_cv <- tidy(cvfit1)
    glance_cv <- glance(cvfit1)

    # plot of MSE as a function of lambda
    g <- ggplot(tidied_cv, aes(lambda, estimate)) + geom_line() + scale_x_log10()
    g

    # plot of MSE as a function of lambda with confidence ribbon
    g <- g + geom_ribbon(aes(ymin = conf.low, ymax = conf.high), alpha = .25)
    g

    # plot of MSE as a function of lambda with confidence ribbon and choices
    # of minimum lambda marked
    g <- g + geom_vline(xintercept = glance_cv$lambda.min) +
        geom_vline(xintercept = glance_cv$lambda.1se, lty = 2)
    g

    # plot of number of zeros for each choice of lambda
    ggplot(tidied_cv, aes(lambda, nzero)) + geom_line() + scale_x_log10()

    # coefficient plot with min lambda shown
    tidied <- tidy(cvfit1$glmnet.fit)
    ggplot(tidied, aes(lambda, estimate, group = term)) + scale_x_log10() +
        geom_line() +
        geom_vline(xintercept = glance_cv$lambda.min) +
        geom_vline(xintercept = glance_cv$lambda.1se, lty = 2)
}