plot_lattice {hmer}

Plot Lattice of Emulator Implausibilities

Description

Plots a set of projections of the full-dimensional input space.

Usage

plot_lattice(
  ems,
  targets,
  ppd = 20,
  cb = FALSE,
  cutoff = 3,
  maxpoints = 50000,
  imp_breaks = NULL,
  contour = TRUE,
  ranges = NULL,
  raster_imp = FALSE,
  plot_vars = NULL,
  fixed_vars = NULL
)

Arguments

ems

The Emulator objects in question.

targets

The corresponding target values.

ppd

The number of points to sample per dimension.

cb

Whether or not a colourblind-friendly plot should be produced.

cutoff

The cutoff value for non-implausible points.

maxpoints

The limit on the number of points to be evaluated.

imp_breaks

If plotting nth maximum implausibility, defines the levels at which to draw contours.

contour

Logical determining whether to plot implausibility contours or not.

ranges

Parameter ranges. If not supplied, defaults to emulator ranges.

raster_imp

Should the implausibility plots be rasterised?

plot_vars

If provided, indicates which subset of parameters to plot.

fixed_vars

If provided, indicates the fixed value of the plot-excluded parameters.

Details

The plots are:

One dimensional optical depth plots (diagonal);

Two dimensional optical depth plots (lower triangle);

Two dimensional minimum implausibility plots (upper triangle).

The optical depth is calculated as follows. A set of points is constructed across the full d-dimensional parameter space, and implausibility is calculated at each point. The points are collected into groups based on their placement in a projection to a one- or two-dimensional slice of the parameter space. For each group, the proportion of non-implausible points is calculated, and this value in [0,1] is plotted. The minimum implausibility plots are similar, but with minimum implausibility calculated rather than proportion of non-implausible points.

The maxpoints argument is used as a cutoff for if a regular ppd grid would result in a very large number of points. If this is the case, then maxpoints points are sampled uniformly from the region instead of regularly spacing them.

If only a subset of parameters are relevant, then the plot_vars and fixed_vars can be used to specify the subset. If plot_vars is provided, corresponding to a list of parameter names, then those parameters not included are fixed to their mid-range values; if fixed_vars is provided as a named list, then the parameters not included are fixed to the corresponding specified values.

Value

A ggplot object

References

Bower, Goldstein & Vernon (2010) <doi:10.1214/10-BA524>

See Also

Other visualisation tools: behaviour_plot(), diagnostic_wrap(), effect_strength(), emulator_plot(), hit_by_wave(), output_plot(), plot_actives(), plot_wrap(), simulator_plot(), space_removed(), validation_pairs(), wave_dependencies(), wave_points(), wave_values()

Examples

 # Excessive runtime
 plot_lattice(SIREmulators$ems, SIREmulators$targets, ppd = 10)
 plot_lattice(SIREmulators$ems$nS, SIREmulators$targets)
 plot_lattice(SIREmulators$ems, SIREmulators$targets, plot_vars = c('aSI', 'aIR'))
 plot_lattice(SIREmulators$ems, SIREmulators$targets, fixed_vars = list(aSR = 0.03))