gw_single_bw_gwr: Evaluate a single GWR bandwidth

In gwverse/gwregr: GWR Package

Evaluate a single GWR bandwidth

Description

This returns a function that evaluates a geographically weighted regression under a given bandwidth. It uses functions from the core gwverse package (gw) to specify local observation and selection and weighting functions (get_neraby_func, gw_get_weight) and requires an evaluation function to be specified (see get_lm_eval_fun in this package). The returned function can be used evaluate different bandwidths, for a given regression model formula.

Usage

gw_single_bw_gwr(sf_data, adaptive, kernel, eval)

Arguments

sf_data	A point or polygon spatial dataset in sf format, containing the attributes to modelled.
kernel	The type of distance weighting to be used: one of "bisquare", "gaussian", "exponential", "tricube" or "boxcar".
eval	An evaluation method for the local model, either "AIC" for a corrected AIC evaluation, or "CV" for a leave one out corss validation.
adative	A logical value TRUE or FALSE to indicate whether an adaptive or fixed bandwidth distance is being used.

Value

The evaluation measure, to be minimised in bandwidth evaluation.

Examples

library(sf)
# load data and define a model formula
data(georgia)
formula = as.formula(MedInc ~ PctBach + PctEld)
## 1. A single ADAPTIVE bandwidth
# create the function
gwr_bw_func = gw_single_bw_gwr(georgia,adaptive=TRUE,kernel="bisquare",eval="AIC")
# apply it
gwr_bw_func(bw = 100, formula)	
## 2. Determining an optimal bandwidth  with `optimse` 
optimise(gwr_bw_func,c(10,nrow(georgia)),formula=formula,maximum=FALSE)
## 3. Determining an adaptive bandwidth with a linear search
# slower but guarantees local minima are not returned
bwsa = 10:nrow(georgia)
# apply the function to all adaptive bandwidths
res = sapply(bwsa, function(x) gwr_bw_func(x, formula))
# find the minimum
bwsa[which.min(res)]
## 4. Determining an optimal FIXED bandwidth 
gwr_bw_func = gw_single_bw_gwr(georgia,adaptive=FALSE,kernel="tricube",eval="CV")
gwr_bw_func(bw =130000, formula)
dist_mat = as.matrix(dist(st_coordinates(st_centroid(georgia)), upper = T, diag = T))
bwsf = seq(50000, (ceiling(max(dist_mat)/5000) * 5000), 5000)
# 4a) by optimising
optimise(gwr_bw_func,bwsf,formula=formula,maximum=FALSE)
# 4b) with a linear search
res = sapply(bwsf, function(x) gwr_bw_func(x, formula))
bwsf[which.min(res)]

gwverse/gwregr documentation built on May 12, 2022, 8:41 p.m.