Admin level rasters (GAUL code as pixel value) containing pixels with large values/GAUL codes cause excessive memory usage. (With GAUL codes reaching ~1M, >32GB of RAM is required)

This is associated with creation of the 'level_all_codes' in 'extractAdmin' which creates an array with size based on the largest pixel value/GAUL code in the raster, rather than by the number of unique pixel values/GAUL codes in the raster.

What does the distance in the nearestLand function mean? Is it the nearest distance of a point to the centroid of the closest raster cell? The coordinates so formed is the centroid so is this the nearest distance that is being measured?

Hello All,
I have a problem with applying the tutorial on this line:
brt <- runBRT(lis$data, 4:6, 1, covariates, gbm.coords = 2:3)

I have this message:

Thanks a lot,
Ahmed

For discrete covariates extractBhatt attempts to run extractAdmin for gaul/polygon occurrences with fun = 'modal'. This argument is passed to library(raster)'s zonal method. 'modal' is not a supported argument for zonal, R's modal (no quotes) function could work, but covariate stacks aligned to the SEEG master grid are too large for this. One option might be to use modal against one covariate layer at a time.

Hello there,

I am trying to install the seegSDM package but I am getting the following error:

devtools::install_github('SEEG-Oxford/seegSDM')
Downloading github repo SEEG-Oxford/seegSDM@master
Error in stri_detect_regex(string, pattern, opts_regex = attr(pattern, :
object 'C_stri_detect_regex' not found

Is the package still available for downloading? If so, could this be an installation bug?

Thanks very much for your assistance.

Mac OS X Yosemite 10.10.5

When factor is of the wrong length extractBhatt should throw a useful error.

I am just leaving this comment here for others that are interested in taking advantage of this package - I appreciate it is not necessarily designed for wider use though. I noticed that seegSDM::nearestLand() assumes planar coordinates. It would be very helpful to add this to the documentation. Alternatively, you could consider tweaking the function to enable correct geodesic distance calculations, e.g.,:

nearestLand2 <- 
  function(points, raster, max_distance, lonlat) {
    nearest <- function(lis, raster) {
      neighbours <- matrix(lis[[1]], ncol = 2)
      point <- lis[[2]]
      land <- !is.na(neighbours[, 2])
      if (!any(land)) {
        return(c(NA, NA))
      }
      else {
        coords <- raster::xyFromCell(raster, neighbours[land, 1])
        if (nrow(coords) == 1) {
          return(coords[1, ])
        }
        # Use raster::pointDistance() or a related function to permit lon/lat distance calculations 
        dists <- raster::pointDistance(coords, cbind(point[1], point[2]), lonlat = lonlat)
        return(coords[which.min(dists), ])
      }
    }
    neighbour_list <- raster::extract(raster, points, buffer = max_distance, 
                                      cellnumbers = TRUE)
    neighbour_list <- lapply(1:nrow(points), function(i) {
      list(neighbours = neighbour_list[[i]], point = as.numeric(points[i, 
      ]))
    })
    return(t(sapply(neighbour_list, nearest, raster)))
  }

dependency on optimal.thresholds

If runBRT was run with model='perf' or model='gbm', getStats fails as the submodels for each fold of the optimal model are not kept.

Maybe this can be fixed by running a single gbm.step starting and ending on n.trees (for method='gbm') or the result of gbm.perf (for method = 'perf')?