Plot a circular-response GAM fit

The default plot method for circ_gam fits: geometry-aware curves that plot.gam does not provide, covering the three regression geometries – circular–linear (cylinder), circular–circular (torus) and linear–circular (the upright can). For a fit with several covariates, no covariate, or an ordinary (non-circlss) family it falls back to mgcv's per-smooth term plots.

Usage

# S3 method for class 'circ_gam'
plot(
  x,
  view = c("flat", "geometry", "both"),
  n = 200,
  se = TRUE,
  pages = 1,
  rug = TRUE,
  ...
)

Arguments

x

A fitted circ_gam model.

view

"flat" (default) draws one panel per distribution parameter on the response scale against the covariate. A circular location or direction carries a band of \(\pm\) one circular standard deviation \(\sqrt{-2\log R}\) of the fitted law (its predictive angular spread, with \(R\) the mean resultant length), wrapped around the \(\pm\pi\) branch cut; a non-circular parameter (the concentration, ...) keeps a delta-method 2-SE band. The location panel is broken at the \(\pm\pi\) branch jump and the observed responses are overlaid. "geometry" draws the fitted curve on its natural surface – a cylinder for circular–linear (the response angle wraps the tube, the covariate runs along the axis), a torus for circular–circular (covariate around the ring), or an upright can for linear–circular (a linear response over a cyclic covariate: the covariate wraps the ring, the response is the height) – chosen from whether the response is circular and whether the covariate is cyclic. "both" places the geometry canvas beside the full set of flat parameter panels – exactly the panels "flat" draws, so the geometry and flat views never disagree.

n

Number of grid points along the covariate.

se

Draw the uncertainty band: \(\pm\) the circular standard deviation for a circular location/direction, a 2-SE interval for a non-circular parameter.

pages

If 1, lay the flat panels out on a single page.

rug

Add a covariate rug to the location panel.

...

Passed to plot.gam in the fallback cases.

Value

The fitted model, invisibly.

Details

The geometry canvas is base-graphics only (persp + trans3d) and is chosen descriptively from the covariate's basis; it is never a fitting input. For pnlss the curve drawn is the derived mean direction \(\mathrm{atan2}(\mu_2, \mu_1)\).

See also

circ_gam, plot.gam