Diagnostic plots for a meep cross-fitted ensemble

Visualises out-of-fold (OOF) ensemble performance per nuisance. For each selected nuisance the panel layout is keyed to that nuisance's family:

• binomial – one ROC panel overlaying each present learner's OOF ROC curve (thin, coloured) plus the combined "stack" curve (thick, black), with a 45-degree reference line and a legend reporting each curve's AUC.

• gaussian – two panels: an OOF \(R^2\) bar chart (one bar per present learner plus a "stack" bar) and an observed-vs-OOF-prediction scatter for the stack with a 45-degree line and the stack \(R^2\) annotated.

Usage

# S3 method for class 'meep'
plot(x, which = c("outcome", "treatment"), ...)

Arguments

x

An object of class "meep", as returned by meep().

which

Character vector of nuisances to draw; any subset of c("outcome", "treatment", "mu0", "mu1"). Nuisances not present on the object are silently dropped. Defaults to c("outcome", "treatment").

...

Currently ignored.

Value

Invisibly returns x.

Details

Panels are laid out on a near-square grid sized to the total panel count (binomial contributes 1 panel, gaussian 2). The function works in headless (non-interactive) mode without error; degenerate nuisances (single-class response, zero-variance outcome) are skipped gracefully rather than erroring.

See also

meep()