R/gwlars.cv.f.r

In wrbrooks/gwselect: Variable selection in Geographically Weighted Regression models

gwlars.cv.f = function(formula, data, weights, bw, N=N, coords, fit.loc, indx, gweight, verbose, adapt, interact, longlat, s, mode.select, tol, method, parallel, precondition, oracle, shrunk.fit, AICc) {    
    #Generate the model with the given bandwidth:
    cat(paste("starting bw:", round(bw, 3), '\n', sep=''))
    gwlars.model = gwlars(formula=formula, data=data, weights=weights, coords=coords, fit.loc=fit.loc, longlat=longlat, gweight=gweight, N=N, bw=bw, indx=indx, tuning=TRUE, predict=FALSE, adapt=adapt, s=s, mode.select=mode.select, method=method, parallel=parallel, precondition=precondition, oracle=oracle, verbose=verbose, interact=interact, shrunk.fit=shrunk.fit, AICc=AICc)

    if (AICc) {
        trH = sum(sapply(gwlars.model[['model']][['models']], function(x) x[['loss.local']])) 

        #Local sigma^2:
        #loss = nrow(data) * (mean(sapply(gwlars.model[['model']][['models']], function(x) log(x[['sigma2']]))) + 1 + (2*(trH+1))/(nrow(data)-trH-2) + log(2*pi))

        #Global sigma^2:
        #if (family=='gaussian') {
        #	#Use the AICc
        	loss = nrow(data) * (log(mean(sapply(gwlars.model[['model']][['models']], function(x) x[['ssr.local']]))) + 1 + (2*(trH+1))/(nrow(data)-trH-2) + log(2*pi))
        #}
        #else if (family %in% c('binomial', 'poisson')) {
        #	#Use GCV (OSullivan et al. 1986)
        #	loss = sum(sapply(gwlars.model[['model']][['models']], function(x) x[['ssr.local']])) / (nrow(data)-trH)**2
        #}
        
        #"Simplistic" BIC - based on eq4.22 from the Fotheringham et al. book:
        #loss = nrow(data) * (log(mean(sapply(gwlars.model[['model']][['models']], function(x) x[['ssr.local']]))) + 1 + log(2*pi)) + trH * log(nrow(data))/2
    }
    else {loss = sum(sapply(gwlars.model[['model']][['models']], function(x) x[['loss.local']]))}

    cat(paste('Bandwidth: ', round(bw, 3), '. Loss: ', round(loss, 3), '\n', sep=''))
    return(loss)
}