Thanks. Truncation is my guess. Another student geocoded the data and wrote it to a CSV file. It's possible that either the site truncated, or the precision when written to CSV or imported into R was modified. The dataset contains observations from a large county in California, so there are sections with very dense sets of observations. It may be possible that the same coordinates appeared multiple times.
I stopped the computation since I used the wrong command, and the following warnings appeared: 1: NA/Inf replaced by maximum positive value in: optimize(gwr.cv.f, lower = beta1, upper = beta2, maximum = FALSE, 2: NA/Inf replaced by maximum positive value in: optimize(gwr.cv.f, lower = beta1, upper = beta2, maximum = FALSE, 3: NA/Inf replaced by maximum positive value in: optimize(gwr.cv.f, lower = beta1, upper = beta2, maximum = FALSE, Three of the CV scores were NA and there were 3 warnings so it looks like R could not optimize the CV curve, but that is the extent of my knowledge. Is this basically just because of coordinates appearing multiple times? Will the resulting adaptive kernel still yield results? (Out of about 15 CV scores it printed, 3 were NA). Thanks, Ryan On 7/1/07, Danlin Yu <[EMAIL PROTECTED]> wrote: > > Ryan: > > 1. The code you are giving actually doesn't use an adaptive kernel. You > need to turn the adpt = True parameter on (its default is False), it > should read like this: > > col.d <- gwr.sel > (ResponseTime~IsRural+IsSemiRural,data=sb,coords=cbind(sb$Longitude,sb$Latitude),adapt > = True, gweight= gwr.bisquare,longlat=TRUE) > > > 2. Without seeing your data, it's hard to see why an "NA" is returned > for your CV score, probably there is something wrong with your > coordinates? Since you mentioned you were using like 19,000 data points, > and using long. and lat. as coordinates. For 19,000 data points, I > assume that they are very close to one another in space (in a city?), > and the long. and lat. are recorded in decimal degrees? If so, one > probability is that when the coordinates were generated by either R or > some other program (ArcGIS, maybe?), due to trunctuation of the > significant decimal points, coordinates for some points that are very > close to one another might have the same values. This could be a > problem. Maybe Roger can give more insights? > > 3. If you don't want to see the results being printed to the screen, you > need to turn the verbose parameter off (its default is True), hence the > code will read like this: > > col.d <- gwr.sel > (ResponseTime~IsRural+IsSemiRural,data=sb,coords=cbind(sb$Longitude,sb$Latitude),adapt > = True, gweight= gwr.bisquare,verbose=False, longlat=TRUE) > > 4. And yes, gwr is very time comsuming (not only gwr.sel). For 19,000 > data points, even using the cross-validation score method, it'll take a > very long time even on a high speed machine. If you are using the AIC > method (which is actually preferred), I am not even sure the code will > run. > > Hope this helps. > > Cheers, > Danlin Yu > > R R R wrote: > > >Hi, > > > >I am trying to use a spatially adaptive kernel in spgwr and I have some > >questions. > > > >I am using the following code: > >col.d <- gwr.sel > > >(ResponseTime~IsRural+IsSemiRural,data=sb,coords=cbind(sb$Longitude,sb$Latitude),gweight= > >gwr.bisquare,longlat=TRUE) > > > >First, I keep getting a CV score of "NA" for each result returned. What > does > >this mean? I do not have any missing values in any observations. Also, > while > >gwr.sel is running, results are printed to the screen, but what does each > >line of ouput represent? Each regression point? > > > >Finally, is gwr.sel expected to take a very long time to complete? I have > >19,000 data points that I am hoping to use for a GW regression but so far > >only two lines of output have been printed to the screen after about 4 > >hours. > > > >I apologize in advance for the naivete of my questions, but I am a newbie > >;-). > > > >Thanks, > >Ryan > > > > [[alternative HTML version deleted]] > > > >_______________________________________________ > >R-sig-Geo mailing list > >R-sig-Geo@stat.math.ethz.ch > >https://stat.ethz.ch/mailman/listinfo/r-sig-geo > > > > > > -- > ___________________________________________ > Danlin Yu, Ph.D. > Assistant Professor > Department of Earth & Environmental Studies > Montclair State University > Montclair, NJ, 07043 > Tel: 973-655-4313 > Fax: 973-655-4072 > email: [EMAIL PROTECTED] > > -- <a href="http://www.spreadfirefox.com/?q=affiliates&id=0&t=180";><img border="0" alt="Get Thunderbird!" title="Get Thunderbird!" src=" http://sfx-images.mozilla.org/affiliates/thunderbird/thunderbird_large.png"/></a> [[alternative HTML version deleted]] _______________________________________________ R-sig-Geo mailing list R-sig-Geo@stat.math.ethz.ch https://stat.ethz.ch/mailman/listinfo/r-sig-geo
![http://sfx-images.mozilla.org/affiliates/thunderbird/thunderbird_large.png"/](http://sfx-images.mozilla.org/affiliates/thunderbird/thunderbird_large.png"/){kind=link}