[R-sig-Geo] spatial clusters

Wed Dec 22 13:16:55 CET 2010

On Wed, 22 Dec 2010, dorina.lazar wrote:

>
> Dear Roger,
>
> Thanks, your suggestions was helpful. I think I exaggerated the questions,
> but for all countries:  “Error in mstree(le, 5): Graph is not
> connected!”; taking only some samples of countries the graph is
> connected. There is something to be done in this case?

If I understand the approach used in SKATER correctly, there may only be 
one graph component, linking all the observations; check with:

n.comp.nb(nb)$nc

which I think must be 1. Otherwise it will not be able to divide the 
(single) graph into clusters, because the input object imposes its own 
subgraphs. Make sure that your samples meet the condition that:

n.comp.nb(nb)$nc == 1

Roger

>
> All the best,
> Dorina
>
>
>
> On Tue, 21 Dec 2010 09:40:35 +0100 (CET), Roger Bivand
> <Roger.Bivand at nhh.no> wrote:
>> On Mon, 20 Dec 2010, dorina.lazar wrote:
>>
>>>
>>> Dear all,
>>>
>>> Finally, with your help, I make some progress to apply SKATER algorithm
>>> to
>>> European countries.
>>>
>>> Here is the R code, following the example from  spdep package pdf:
>>>
>>>> country<-wrld_simpl[wrld_simpl$ISO3%in%c("AZE", "ALB", "ARM", "BIH",
>>> "BGR", "CYP", "DNK", "IRL", "EST", "AUT", "CZE", "FIN",  "FRA", "GEO",
>>> "DEU", "GRC", "HRV", "HUN", "ISL", "ITA", "LVA", "BLR", "LTU", "SVK",
>>> "LIE", "MLT", "BEL", "AND",  "LUX", "MCO", "NLD", "NOR", "POL", "PRT",
>>> "ROU", "MDA", "RUS", "SVN"),]
>>>> ske<-spCbind(country, rclasif)
>>>> ske at data
>>>> de <- data.frame(scale(ske at data[13]))   #col 13 contains the health
>>> indicator (from rclasif)
>>>> nbe <- poly2nb(country)
>>>> ce <- nbcosts(nbe, de)
>>>> le <- nb2listw(nbe, ce, style="B", zero.policy=TRUE)
>>> Error in nb2listw(nbe, ce, style = "B", zero.policy = TRUE) :
>>> neighbours and glist do not conform
>>
>> Using your corrected list of countries, I see two/three problems:
>>
>> 1) some proximate polygons do not touch because of line generalisation
> in
>> the simplification of the country polygons, use:
>>
>> nbe <- poly2nb(country, snap=0.1)
>>
>> to force neighbours to touch.
>>
>> 2) you are left with three real islands - Iceland, Cyprus, and Malta. To
>
>> join them in, consider:
>>
>> knn4 <- knn2nb(knearneigh(crds, 4, longlat=TRUE),
>>    row.names=row.names(country))
>> knn4s <- make.sym.nb(knn4)
>> nbe1 <- union.nb(nbe, knn4s)
>>
>> 3) nbcosts() does not like single-column data frames, as it is supposed
> to
>> be for multivariate data, so:
>>
>> de <- data.frame(rnorm(46))
>> ce <- nbcosts(nbe1, de)
>> which(sapply(ce, function(x) isTRUE(any(is.na(x)))))
>>
>> gives an NA at the end of each line of cost distances, but:
>>
>> de <- data.frame(rnorm(46), rnorm(46))
>> ce <- nbcosts(nbe1, de)
>> which(sapply(ce, function(x) isTRUE(any(is.na(x)))))
>> le <- nb2listw(nbe1, ce, style="B")
>>
>> doesn't, nor does:
>>
>> de <- data.frame(rnorm(46), rep(1, 46))
>> ce <- nbcosts(nbe1, de)
>> which(sapply(ce, function(x) isTRUE(any(is.na(x)))))
>> le <- nb2listw(nbe1, ce, style="B")
>>
>> From there,
>>
>> mst.bh <- mstree(le,5)
>> res1 <- skater(mst.bh[,1:2], de, 2)
>> plot(country, col=heat.colors(3)[res1$groups])
>>
>> seem to work.
>>
>> I think that it makes sense for nbcosts() to fail for non-connected
>> observations, but perhaps it should explain why.
>>
>> Hope this helps,
>>
>> Roger
>>
>>
>>>
>>> I mention that:
>>> -	at "nbe", appears the message:  6 regions with no links: CYP DNK HRV
>>> ISL
>>> MLT SWE
>>> -	"ce" contain a lot of NA
>>> If you have some ideas about the cause of this Error, please help me to
>>> finalize my exercise on SKATER algoritm.
>>>
>>> Thanks a lot for your help,
>>> Dorina
>>>
>>>
>>>
>>>
>>> On Sat, 18 Dec 2010 18:42:35 +0100 (CET), Roger Bivand
>>> <Roger.Bivand at nhh.no> wrote:
>>>> On Sat, 18 Dec 2010, Roman Luštrik wrote:
>>>>
>>>>> Perhaps you could follow the instructions from this
>>>>>
>>>
> exchange<http://r-sig-geo.2731867.n2.nabble.com/Guam-island-map-td4969467.html#a4972725>and
>>>>> see if you can extract the desired country (assuming the resolution
>>>>> suits you).
>>>>>
>>>>
>>>> I don't think that we know which countries Dorina needs, or at which
>>>> administrative level. If there are only 30 country observations, and
> the
>>>
>>>> countries are not contiguous, then the whole exercise seems
> unnecessary.
>>>
>>>> Can Dorina use the simplified world map in the maptools package:
>>>>
>>>> library(maptools)
>>>> ?wrld_simpl
>>>>
>>>> adding the data using the ISO country names? If these are countries at
>>>> some administrative division level, then maybe www.gadm.org will have
>>> what
>>>> is needed, but the countries will have to be joined together, or the
>>> whole
>>>> map subsetted. If these are say EU NUTS2 regions, then use can be made
>>> of
>>>> GISCO shapefiles from Eurostat. But we don't know what boundaries are
>>>> needed, and of course to use Skater for clustering, you need the
>>>> boundaries.
>>>>
>>>> Roger
>>>>
>>>>>
>>>>> Cheers,
>>>>> Roman
>>>>>
>>>>>
>>>>>
>>>>> On Sat, Dec 18, 2010 at 6:20 PM, Dorina Lazar <drnlazar at yahoo.com>
>>> wrote:
>>>>>
>>>>>> Georg,
>>>>>>
>>>>>> Thanks for being very patient with me.
>>>>>>
>>>>>> I have not a shapefile for countries.
>>>>>>
>>>>>> I have tried to obtain it. I have loaded a shapefile for Finland,
> GIS
>>>>>> data,
>>>>>> from
>>>>>>
>>>
> http://www.eea.europa.eu/data-and-maps/data/eea-reference-grids/zipped-shape-file-finland
>>>>>>
>>>>>> and I tried to read the file FI_1K.shp
>>>>>> Fin<- readShapePoly(system.file("etc/shapes/ FI_1K.shp",
>>>>>> package="spdep"))
>>>>>> but...
>>>>>> Error in getinfo.shape(filen) : Error opening SHP file
>>>>>>
>>>>>> How can I obtain, for each country,  the shapefile spatial polygon
> in
>>> R
>>>>>> (SpatialPolygons-DataFrame)?
>>>>>>
>>>>>> Thanks again,
>>>>>> Dorina
>>>>>>
>>>>>>
>>>>>> --- On Thu, 12/16/10, Georg Ru? <research at georgruss.de> wrote:
>>>>>>
>>>>>>> From: Georg Ru? <research at georgruss.de>
>>>>>>> Subject: Re: [R-sig-Geo] spatial clusters
>>>>>>> To: "Dorina Lazar" <dorina.lazar at econ.ubbcluj.ro>
>>>>>>> Cc: "R-sig-geo mailing list" <r-sig-geo at r-project.org>
>>>>>>> Date: Thursday, December 16, 2010, 10:41 AM
>>>>>>> On 16/12/10 09:06:33, dorina.lazar
>>>>>>> wrote:
>>>>>>>> How to create a shapefile including my data about some
>>>>>>> socio-demographic
>>>>>>>> indicators (7),  for about 30 countries around
>>>>>>> the world (to be read with
>>>>>>>> readShapePoly)?
>>>>>>>
>>>>>>> Hi Dorina,
>>>>>>>
>>>>>>> I'm not quite sure what you have and what you want.
>>>>>>>
>>>>>>> I assume that you'd like to create a shapefile such that
>>>>>>> for every country
>>>>>>> (which is represented as a polygon) there's a vector of
>>>>>>> seven attributes
>>>>>>> attached to it:
>>>>>>>
>>>>>>> country1:  attr1 attr2 attr3 attr4 attr5 attr6 attr7
>>>>>>> country2:  attr1 attr2 attr3 attr4 attr5 attr6 attr7
>>>>>>> ...
>>>>>>>
>>>>>>> Do you have your countries as spatial polygons in R
>>>>>>> (SpatialPolygons-
>>>>>>> DataFrame or similar)? If yes, you should be able to add
>>>>>>> your socio-demo-
>>>>>>> graphic indicators to this data frame (just as in normal
>>>>>>> data frames). In
>>>>>>> this way you'd join the data you have inside R.
>>>>>>>
>>>>>>> If you have the shapefile for the countries available and
>>>>>>> would like to
>>>>>>> edit this directly to add the indicators, I'd suggest using
>>>>>>> a GIS like
>>>>>>> GRASS http://grass.fbk.eu/ Afterwards, this can be easily
>>>>>>> read into R.
>>>>>>>
>>>>>>>> Would be useful to suggest some good introductory
>>>>>>> lectures in spatial
>>>>>>>> statistics (links or books).
>>>>>>>
>>>>>>> I think the following is what Roger would suggest:
>>>>>>>
>>>>>>>  author = {Bivand, Roger S. and Pebesma, Edzer J. and
>>>>>>> G?mez-Rubio, Virgilio},
>>>>>>>  title = {Applied Spatial Data Analysis with R},
>>>>>>>  series = {Use R},
>>>>>>>  publisher = {Springer},
>>>>>>>
>>>>>>> http://www.asdar-book.org/
>>>>>>> http://www.springerlink.com/content/uw07v1/
>>>>>>> (Depending on your login/institute you may have fulltext
>>>>>>> access, but the
>>>>>>> book is definitely worth buying anyway.)
>>>>>>>
>>>>>>> Regards,
>>>>>>> Georg.
>>>>>>> --
>>>>>>> Research Assistant
>>>>>>> Otto-von-Guericke-Universit?t Magdeburg
>>>>>>> research at georgruss.de
>>>>>>> http://research.georgruss.de
>>>>>>>
>>>>>>> _______________________________________________
>>>>>>> R-sig-Geo mailing list
>>>>>>> R-sig-Geo at r-project.org
>>>>>>> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> _______________________________________________
>>>>>> R-sig-Geo mailing list
>>>>>> R-sig-Geo at r-project.org
>>>>>> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>
>

-- 
Roger Bivand
Economic Geography Section, Department of Economics, Norwegian School of
Economics and Business Administration, Helleveien 30, N-5045 Bergen,
Norway. voice: +47 55 95 93 55; fax +47 55 95 95 43
e-mail: Roger.Bivand at nhh.no