Ani,
The raster::predict function has na.rm=TRUE and removes the NA values, but
not -Inf/Inf values and that indeed seems to be the problem here (see
example below). I have added an option "inf.rm" to the predict function of
the development version of 'raster' (version 2.0-20) you can try to install
it from R-Forge in 12 hours: *install.packages("raster", repos="
http://R-Forge.R-project.org";).*
With that version, your code will likely work if you do predict( ,
inf.rm=TRUE)
Alternatively, you could write your rasters to a new file using
writeRaster, because that will likely replace the -Inf/Inf values with NA,
and then use a RasterBrick made from this new file in the predict version.
When asking questions like this, please try to provide a self-contained
example like I provide below (mostly copied from ?raster::predict).
library(raster)
library(caret)
logo <- brick(system.file("external/rlogo.grd", package="raster"))
#get presence and random-absence points
p <- matrix(c(48, 48, 48, 53, 50, 46, 54, 70, 84, 85, 74, 84, 95, 85,
66, 42, 26, 4, 19, 17, 7, 14, 26, 29, 39, 45, 51, 56, 46, 38, 31,
22, 34, 60, 70, 73, 63, 46, 43, 28), ncol=2)
a <- cbind(runif(250)*(xmax(logo)-xmin(logo))+xmin(logo),
runif(250)*(ymax(logo)-ymin(logo))+ymin(logo))
#extract values for points
xy <- rbind(cbind(1, p), cbind(0, a))
train <- data.frame(extract(logo, xy[,2:3]))
class <- as.factor(xy[,1])
f <- class~red+green+blue
rfFit <- train(as.data.frame(train), class, method = "rf", tuneLength = 5,
trControl = trainControl(method = "repeatedcv", repeats = 5, verboseIter =
FALSE), metric = "Kappa")
> logo[2:10,] <- NA
> r1 <- predict(logo, rfFit)
>
> logo[2:10,] <- Inf
> r2 <- predict(logo, rfFit)
Error in predict.randomForest(modelFit, newdata) :
NA/NaN/Inf in foreign function call (arg 7)
>
Best, Robert
On Wed, Oct 3, 2012 at 7:23 AM, aniruddha ghosh <aniru...@gmail.com> wrote:
> Dear List,
> I'm trying to classify a multiband raster with point shapefiles as
> training data. I used the following:
> vec<-readOGR(".","training")
> raster<-stack("test.tif") #it contains around 100000 pixels and 240
> layers
> outImage<-"classified.tif"
> #then extract the training values
> train<-extract(raster,vec)
> #after that used "caret" package to train the model
> rfFit <- train(as.data.frame(train),as.factor(vec$class),
> method = "rf", tuneLength = 5,
> trControl = trainControl(method = "repeatedcv",
> repeats = 5,
> verboseIter = FALSE),
> metric = "Kappa")
> # then to classify the entire data
> rfPred <- predict(raster,rfFit, filename=outImage,na.rm=TRUE,
> progress='text', format='GTiff', datatype='INT1U', overwrite=TRUE)
>
> after some progress (more than 20 %) I'm getting this error:
> "NA/NaN/Inf in foreign function call (arg 7)"
>
> I have checked the code for small test image (around 10000 pixels with
> 125 layers) and it worked fine. I suspect it is due to the "-Inf"
> present in the large raster. How to deal with this?
>
> Thanks,
> --
> Ani
>
> _______________________________________________
> R-sig-Geo mailing list
> R-sig-Geo@r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>
[[alternative HTML version deleted]]
_______________________________________________
R-sig-Geo mailing list
R-sig-Geo@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-geo
