Skip to contents

Boosting error trajectory data object

Source: R/gg_boost_error.R
gg_boost_error.Rd

Extract the error path of a boosted multivariate tree fit as a function of the boosting iteration, together with the optimal iteration selected by cross-validation.

Usage

gg_boost_error(object, use.rmse = TRUE, ...)

Arguments

object

A fitted boostmtree object, or a fitted BoostMLR object.

use.rmse

Logical. When TRUE (default) return the standardized l2 error; when FALSE return the unstandardized squared error.

...

Not used; present for S3 consistency.

Value

A gg_boost_error data.frame with columns:

iteration

Integer boosting iteration, from 1.

value

Numeric error at that iteration.

response

Factor naming the response.

optimal

Logical, TRUE at the cross-validated optimal iteration.

Details

The fit must have been grown with cv.flag = TRUE. boostmtree records err.rate and m.opt only when cross-validation ran; a default fit carries neither, and there is no error path to extract. If this function reports that the fit has no error path, refit with cv.flag = TRUE.

boostmtree stores the error on the standardized response scale in the l2 column. use.rmse = FALSE returns (l2 * y.sd)^2, the squared error on the original scale.

This figure also accepts a BoostMLR grow fit (predict objects are not supported – see below). BoostMLR records Error_Rate as an M-by-response matrix already on a single scale, so there is no use.rmse argument for that backend; use.rmse = FALSE is refused, while TRUE, the default, names the scale it already returns. BoostMLR grow objects also select no optimal iteration anywhere in the object – partial.BoostMLR() takes Mopt as a user-supplied argument instead – so optimal is FALSE for every row; deriving an argmin here would report a choice the backend never made.

A BoostMLR predict object is refused outright, even though it shares the "BoostMLR" class and also carries an Error_Rate. Its Error_Rate is test error rather than the training path documented above, and it records a real Mopt that this extractor would otherwise silently discard and overwrite with optimal = FALSE.

See also

plot.gg_boost_error, gg_boost_path

Examples

# \donttest{
sim <- boostmtree::simLong(n = 25, n.time = 4, model = 1)$data.list
fit <- boostmtree::boostmtree(
  x = sim$features, tm = sim$time, id = sim$id, y = sim$y,
  M = 50, cv.flag = TRUE, verbose = FALSE
)
gg_dta <- gg_boost_error(fit)
plot(gg_dta)

# }