# plot.tgp: Plotting for Treed Gaussian Process Models In tgp: Bayesian Treed Gaussian Process Models

## Description

A generic function for plotting of `"tgp"`-class objects. 1-d posterior mean and error plots, 2-d posterior mean and error image and perspective plots, and 3+-dimensional mean and error image and perspective plots are supported via projection and slicing.

## Usage

 ```1 2 3 4 5 6 7``` ```## S3 method for class 'tgp' plot(x, pparts = TRUE, proj = NULL, slice = NULL, map = NULL, as = NULL, center = "mean", layout = "both", main = NULL, xlab = NULL, ylab = NULL, zlab = NULL, pc = "pc", gridlen = c(40,40), span = 0.1, pXX = TRUE, legendloc = "topright", maineff = TRUE, mrlayout="both", rankmax = 20, ...) ```

## Arguments

 `x` `"tgp"`-class object that is the output of one of the `b*` functions: `blm`, `btlm` `bgp`, `bgpllm`, `btgp`, or `btgpllm` `pparts` If `TRUE`, partition-regions are plotted (default), otherwise they are not `proj` 1-or-2-Vector describing the dimensions to be shown in a projection. The argument is ignored for 1-d data, i.e., if ```x\$d == 1```. For 2-d data, no projection needs be specified— the default argument (`proj = NULL`) will result in a 2-d perspective or image plot. 1-d projections of 2-d or higher data are are supported, e.g., `proj = c(2)` would show the second variable projection. For 3-d data or higher, `proj=NULL` defaults to `proj = c(1,2)` which plots a 2-d projection for the first two variables. Slices have priority over the projections— see next argument (`slice`)— when non-null arguments are provided for both. `slice` `list` object with `x` and `z` fields, which are vectors of equal length describing the slice to be plotted, i.e., which z-values of the `x\$d - 2` inputs `x\$X` and `x\$XX` should be fixed to in order to obtain a 2-d visualization. For example, for 4-d data, `slice = list(x=(2,4), z=c(0.2, 1.5)` will result in a 2-d plot of the first and third dimensions which have the second and fourth slice fixed at 0.5 and 1.5. The default is `NULL`, yielding to the `proj` argument. Argument is ignored for 1-d data, i.e., if `x\$d == 1` `map` Optional 2-d map (longitude and latitude) from maps to be shown on top of image plots `center` Default `center = "mean"` causes the posterior predictive mean to be plotted as the centering statistic. Otherwise the median can be used with `center = "med"`, or the kriging mean with `center = "km"` `as` Optional string indicator for plotting of adaptive sampling statistics: specifying `as = "alm"` for ALM, `as = "s2"` for predictive variance, `as = "ks2"` for expected kriging variance, `as = "alc"` for ALC, and `as = "improv"` for expected improvement (about the minimum, see the `rankmax` argument below). The default `as = NULL` plots error-bars (1d-plots) or error magnitudes (2d-plots), which is essentially the same as `as = "alm"` `layout` Specify whether to plot the mean predictive surface (`layout = "surf"`), the error or adaptive sampling statistics (`layout = "as"`), or default (`layout = "both"`) which shows both. If `layout = "sens"`, plot the results of a sensitivity analysis (see `sens`) in a format determined by the argument `maineff` below. `main` Optional `character` string to add to the main title of the plot `xlab` Optional `character` string to add to the x label of the plots `ylab` Optional `character` string to add to the y label of the plots `zlab` Optional `character` string to add to the z label of the plots; ignored unless `pc = "p"` `pc` Selects perspective-posterior mean and image-error plots (`pc = "pc"`, the default) or a double–image plot (```pc = "c"```)

(only valid for 2-d plots)

 `gridlen` Number of regular grid points for 2-d slices and projections in x and y. The default of `gridlen = c(40,40)` causes a `40 * 40` grid of `X`, `Y`, and `Z` values to be computed. Ignored for 1-d plots and projections `span` Span for `loess` kernel. The tgp package default (```span = 0.1```) is set lower than the `loess` default. Smaller spans can lead to warnings from `loess` when the data or predictive locations are sparse and ugly plots may result. In this case, try increasing the span `pXX` scalar logical indicating if `XX` locations should be plotted `legendloc` Location of the `legend` included in the plots of sensitivity analyses produced with `layout = "sens"`, or 1-d plots of multi-resolution models (with `corr = "mrexpsep"`) and option `mrlayout = "both"`; otherwise the argument is ignored `maineff` Format for the plots of sensitivity analyses produced with `layout = "sens"`; otherwise the argument is ignored. If `maineff=TRUE` main effect plots are produced alongside boxplots for posterior samples of the sensitivity indices, and if `FALSE` only the boxplots are produced. Alternatively, `maineff` can be a matrix containing input dimensions in the configuration that the corresponding main effects are to be plotted; that is, `mfrow=dim(maineff)`. In this case, a 90 percent interval is plotted with each main effect and the sensitivity index boxplots are not plotted. `mrlayout` The plot layout for double resolution tgp objects with `params\$corr == "mrexpsep"`. For the default `mrlayout="both"`, the coarse and fine fidelity are plotted together, either on the same plot for 1D inputs or through side-by-side image plots of the predicted `center` with axis determined by `proj` for inputs of greater dimension. Note that many of the standard arguments – such as `slice`, `pc`, and `map` – are either non-applicable or unsupported for `mrlayout="both"`. If `mrlayout="coarse"` or `mrlayout="fine"`, prediction for the respective fidelity is plotted as usual and all of the standard options apply. `rankmax` When `as = "improv"` is provided, the posterior expected improvements are plotted according the the first column of the `improv` field of the `"tgp"`-class object. Text is added to the plot near the `XX` positions of the first `1:rankmax` predictive locations with the highest ranks in the second column of the `improv` field. `...` Extra arguments to 1-d (`plot`) and 2-d plotting functions `persp` and `image`

## Value

The only output of this function is beautiful plots

## Note

This plotting function is provided with the intention that it will be used as an aid in the visualization of `"tgp"`-class objects. Users are encouraged to use the source code for this function in order to develop custom plotting functions.

1-d projections for 3-d or higher data are also available by specifying a 1-d projection argument (e.g. `proj=c(1)` for projecting onto the first input variable).

For examples, see `vignette("tgp")` and the help files of those functions in "See Also", below

## Author(s)

Robert B. Gramacy, rbg@vt.edu, and Matt Taddy, mataddy@amazon.com

## See Also

`plot`, `bgpllm`, `btlm`, `blm`, `bgp`, `btgpllm`, `predict.tgp`, `tgp.trees`, `mapT`, `loess`, `sens`

tgp documentation built on Jan. 13, 2021, 3:49 p.m.