Become an expert in R — Interactive courses, Cheat Sheets, certificates and more!
Get Started for Free

predictSurface

Evaluates a fitted function or the prediction error as a surface that is suitable for plotting with the image, persp, or contour functions.


Description

Evaluates a a fitted model or the prediction error on a 2-D grid keeping any other variables constant. The resulting object is suitable for use with functions for viewing 3-d surfaces.

Usage

## Default S3 method:
predictSurface(object, grid.list = NULL, 
                     extrap = FALSE, chull.mask = NA, nx = 80, ny = 80,
                        xy = c(1,2),  verbose = FALSE, ...)

## S3 method for class 'fastTps'
predictSurface(object, grid.list = NULL, 
       extrap = FALSE, chull.mask = NA, nx = 80, ny = 80,
       xy = c(1,2),  verbose = FALSE, ...)
       
## S3 method for class 'Krig'
predictSurface(object, grid.list = NULL, extrap = FALSE, chull.mask = NA, 
nx = 80, ny = 80, xy = c(1, 2), verbose = FALSE, ZGrid = NULL, 
    drop.Z = FALSE, just.fixed=FALSE,  ...)
    
## S3 method for class 'mKrig'
predictSurface(object,  ...)

## Default S3 method:
predictSurfaceSE( object, grid.list = NULL, extrap =
FALSE, chull.mask = NA, nx = 80, ny = 80, xy = c(1,2), verbose =
FALSE, ...)

## S3 method for class 'surface'
predict(object,...)

Arguments

object

An object from fitting a function to data. In fields this is usually a Krig, mKrig, or fastTps object.

grid.list

A list with as many components as variables describing the surface. All components should have a single value except the two that give the grid points for evaluation. If the matrix or data frame has column names, these must appear in the grid list. See the grid.list help file for more details. If this is omitted and the fit just depends on two variables the grid will be made from the ranges of the observed variables. (See the function fields.x.to.grid.)

extrap

Extrapolation beyond the range of the data. If FALSE (the default) the predictions will be restricted to the convex hull of the observed data or the convex hull defined from the points from the argument chull.mask. This function may be slightly faster if this logical is set to TRUE to avoid checking the grid points for membership in the convex hull. For more complicated masking a low level creation of a bounding polygon and testing for membership with in.poly may be useful.

chull.mask

Whether to restrict the fitted surface to be on a convex hull, NA's are assigned to values outside the convex hull. chull.mask should be a sequence of points defining a convex hull. Default is to form the convex hull from the observations if this argument is missing (and extrap is false).

nx

Number of grid points in X axis.

ny

Number of grid points in Y axis.

xy

A two element vector giving the positions for the "X" and "Y" variables for the surface. The positions refer to the columns of the x matrix used to define the multidimensional surface. This argument is provided in lieu of generating the grid list. If a 4 dimensional surface is fit to data then xy= c(2,4) will evaluate a surface using the second and fourth variables with variables 1 and 3 fixed at their median values. NOTE: this argument is ignored if a grid.list argument is passed.

drop.Z

If TRUE the fixed part of model depending on covariates is omitted.

just.fixed

If TRUE the nonparametric surface is omitted.

...

Any other arguments to pass to the predict function associated with the fit object. Some of the usual arguments for several of the fields fitted objects include:

ynew

New values of y used to reestimate the surface.

Z

A matrix of covariates for the fixed part of model.

ZGrid

An array or list form of covariates to use for prediction. This must match the grid.list argument. e.g. ZGrid and grid.list describe the same grid. If ZGrid is an array then the first two indices are the x and y locations in the grid. The third index, if present, indexes the covariates. e.g. For evaluation on a 10X15 grid and with 2 covariates. dim( ZGrid) == c(10,15, 2). If ZGrid is a list then the components x and y shold match those of grid.list and the z component follows the shape described above for the no list case.

verbose

If TRUE prints out some imtermediate results for debugging.

Details

This function creates the right grid using the grid.list information or the attribute in xg, calls the predict function for the object with these points and also adding any extra arguments passed in the ... section, and then reforms the results as a surface object (as.surface). To determine the what parts of the prediction grid are in the convex hull of the data the function in.poly is used. The argument inflation in this function is used to include a small margin around the outside of the polygon so that point on convex hull are included. This potentially confusing modification is to prevent excluding grid points that fall exactly on the ranges of the data. Also note that as written there is no computational savings for evaluting only the convex subset compared to the full grid.

predictSurface.fastTps is a specific version ( m=2, and k=2) that can be much more efficient because it takes advantage of a low level FORTRAN call to evaluate the Wendland covariance function. Use predictSurface or predict for other choices of m and k.

predictSurface.Krig is designed to also include covariates for the fixed in terms of grids. Due to similarity in output and the model. predictSurface.mKrig just uses the Krig method.

NOTE: predict.surface has been depreciated and just prints out a warning when called.

Value

The usual list components for making contour and perspective plots (x,y,z) along with labels for the x and y variables. For predictSurface.derivative the component z is a three dimensional array with nx, ny, 2.

See Also

Tps, Krig, predict, grid.list, make.surface.grid, as.surface, surface, in.poly

Examples

fit<- Tps( BD[,1:4], BD$lnya)  # fit surface to data 

# evaluate fitted surface for  first two 
# variables holding other two fixed at median values

out.p<- predictSurface(fit)
surface(out.p, type="C") 

#
# plot surface for second and fourth variables 
# on specific grid. 

glist<- list( KCL=29.77, MgCl2= seq(3,7,,25), KPO4=32.13, 
                     dNTP=seq( 250,1500,,25))

out.p<- predictSurface(fit, glist)
surface(out.p, type="C")

out.p<- predictSurfaceSE(fit, glist)
surface(out.p, type="C")

fields

Tools for Spatial Data

v11.6
GPL (>= 2)
Authors
Douglas Nychka [aut, cre], Reinhard Furrer [aut], John Paige [aut], Stephan Sain [aut], Florian Gerber [aut], Matthew Iverson [aut], University Corporation for Atmospheric Research [cph]
Initial release
2020-10-06

We don't support your browser anymore

Please choose more modern alternatives, such as Google Chrome or Mozilla Firefox.