[R-sig-Geo] calculate CRPS on rasters

[Lorenzo Alfieri]

Robert,
you're right, it works. 
I must have done something wrong yesterday
I'm trying to optimize the computation time
I've tested the two alternatives on a predictor of 51 layers of 680x810 pixels

CRPS[] <- crps(obs[],pred[])$crps                 took 0.98 second
CRPS  <- overlay(obs, pred, fun=function(x,y) crps(x,y)$crps)          took 1.45 seconds

anyway, thank you all!

Lorenzo


________________________________
> Date: Thu, 29 Nov 2012 10:34:36 -0800 
> Subject: Re: [R-sig-Geo] calculate CRPS on rasters 
> From: r.hijmans at gmail.com 
> To: alfios17 at hotmail.com 
> CC: r-sig-geo at r-project.org 
>  
> Lorenzo, 
>  
> The below (your example) works for me, using the overlay as I suggested. 
> Perhaps you did not load the verification package? 
>  
> Robert 
>  
> library(verification)      #includes function crps 
> library(raster) 
> set.seed(1) 
> r <- raster(ncol=10, nrow=10) 
> pred1 <- init(r, fun=runif) 
> pred2 <- init(r, fun=runif) 
> pred3 <- init(r, fun=runif) 
> pred <- stack(pred1, pred2, pred3) 
> obs <- init(r, fun=runif) 
>  
> CRPS  <- overlay(obs, pred, fun=function(x,y) crps(x,y)$crps) 
>  
>  
>  > CRPS 
> class       : RasterLayer 
> dimensions  : 10, 10, 100  (nrow, ncol, ncell) 
> resolution  : 36, 18  (x, y) 
> extent      : -180, 180, -90, 90  (xmin, xmax, ymin, ymax) 
> coord. ref. : +proj=longlat +datum=WGS84 
> data source : in memory 
> names       : layer 
> values      : 0.003277372, 0.7726679  (min, max) 
>  
>  
> # More fancy, to get all the crps output: 
>  
> x  <- overlay(obs, pred, fun=function(x,y) as.matrix(data.frame(crps(x,y)))) 
>  
>  
>  
> On Thu, Nov 29, 2012 at 1:15 AM, Lorenzo Alfieri  
> <alfios17 at hotmail.com<mailto:alfios17 at hotmail.com>> wrote: 
>  
> Hi Robert, 
> I was hoping to use the overlay function too, but it seems it doesn't 
> work in this case. See the result: 
>  
> CRPS  <- overlay(obs, pred, fun=function(x,y) crps(x,y)$crps) 
>  
> Error in function (x, fun, filename = "", recycle = TRUE, ...)  : 
>    cannot use this formula, probably because it is not vectorized 
>  
> Lorenzo 
>  
>  
>  
> Date: Wed, 28 Nov 2012 11:25:36 -0800 
> Subject: Re: [R-sig-Geo] calculate CRPS on rasters 
> From: r.hijmans at gmail.com<mailto:r.hijmans at gmail.com> 
> To: alfios17 at hotmail.com<mailto:alfios17 at hotmail.com> 
> CC: etiennebr at gmail.com<mailto:etiennebr at gmail.com>;  
> r-sig-geo at r-project.org<mailto:r-sig-geo at r-project.org> 
>  
> Lorenzo, 
> Given that solution, I think you can also express this (in a  
> memory-safe fashion) like this : 
> CRPS  <- overlay(obs, pred, fun=function(x,y) crps(x, y)$crps)
>  
> Robert 
>  
> On Tue, Nov 27, 2012 at 3:11 AM, Lorenzo Alfieri  
> <alfios17 at hotmail.com<mailto:alfios17 at hotmail.com>> wrote: 
>  
>  
>  
>  
> Etienne, 
>  
> thank you for the tip 
>  
> Now it runs trough, by using 
>  
>  
>  
> CRPS <- raster(ncol=10, nrow=10) 
>  
> CRPS[] <- crps(obs[],pred[])$crps 
>  
>  
>  
> Lorenzo 
>  
>  
>  
>  
>  
> Date: Mon, 26 Nov 2012 21:37:13 -0500 
>  
> Subject: Re: [R-sig-Geo] calculate CRPS on rasters 
>  
> From: etiennebr at gmail.com<mailto:etiennebr at gmail.com> 
>  
> To: alfios17 at hotmail.com<mailto:alfios17 at hotmail.com> 
>  
> CC: r-sig-geo at r-project.org<mailto:r-sig-geo at r-project.org> 
>  
>  
>  
> Lorenzo, 
>  
>  
>  
> I don't know about your specific function, but you can access matrix  
> using the []'s. So maybe something like : 
>  
> CRPS <- raster(ncol=10, nrow=10) 
>  
> CRPS[] <- crps(obs[],pred[]) 
>  
>  
>  
>  
>  
> Etienne 
>  
>  
>  
> 2012/11/26 Lorenzo Alfieri  
> <alfios17 at hotmail.com<mailto:alfios17 at hotmail.com>> 
>  
>  
>  
> CRPS <- raster(ncol=10, nrow=10) 
>  
>  
>  
> for (i in 1:length(obs)){ 
>  
>  
>  
>    CRPS[i] <- crps(obs[i],pred[i])$CRPS 
>  
>  
>  
>          [[alternative HTML version deleted]] 
>  
>  
>  
> _______________________________________________ 
>  
> R-sig-Geo mailing list 
>  
> R-sig-Geo at r-project.org<mailto:R-sig-Geo at r-project.org> 
>  
> https://stat.ethz.ch/mailman/listinfo/r-sig-geo 
>  
>  
>  
>