Michael Sumner's rasters tips about geocompr HOT 8 CLOSED

Thanks for \4... this seems still to improve performance significantly.

from geocompr.

raster tips: set options(rasterMaxMemory = nbytes) to change limit for redirect to file (by default not set, hard coded to 100Mb)

https://twitter.com/mdsumner/status/868655402224762880
https://gist.github.com/mdsumner/7cb21ce030ca5e9acae7d14e0d929162

from geocompr.

@mdsumner thanks for providing this list. Are they still your top tips or would you make any alterations to them based on recent developments? Are any not futureproof?

from geocompr.

4 and 5 are bugs, so they could be fixed - other than that I think they're pretty good :)

I'll add "use spex::polygonize(raster) for fastest conversion to a polygon-per-pixel sf layer /14"

for 2 I'd add use tabularaster::cellnumbers instead of juggling lists of cells

raster does now have some sf support (for extract, by conversion to Spatial)

also watch rhijmans/geospat, r-spatial/stars, and ecohealthalliance/fasterize

from geocompr.

Great - I think we've got most of those covered now:

• General intro to raster: http://geocompr.robinlovelace.net/spatial-class.html#raster-data
• Spatial operations: http://geocompr.robinlovelace.net/spatial-operations.html#spatial-ras
• Geometric operations: http://geocompr.robinlovelace.net/transform.html#geo-ras
• Raster-vector: http://geocompr.robinlovelace.net/raster-vector.html

One question: is the distinction between spatial and geometric operations clear? And have we missed anything? Thanks @mdsumner for all these tips, hope we've done them justice!

from geocompr.

LGTM! The book is awesome btw, it's great to see this all laid out in detail.

5/ is now fixed, it's in the queue on CRAN

There's no need to as(x, "Spatial") for extract now for sf, since Nov 2017 (support for sf is patchy but more is coming)

One new "tip" is that extract can be massively sped up by using fasterize, but it's a bit abstract - and presumably raster will apply this improvement some time soon. I also didn't realize extract had that df argument (which is embarassing).

library(raster)
library(sf)

#install.packages("spDataLarge", repos = "https://nowosad.github.io/drat/",
#                 type = "source")
library(spDataLarge)
srtm = raster(system.file("raster/srtm.tif", package = "spDataLarge"))
zion = read_sf(system.file("vector/zion.gpkg", package = "spDataLarge"))
zion = st_transform(zion, projection(srtm))

fast_extract <- function(x, y) {
  y[["ID"]] <- seq_len(nrow(y))
  rast_poly <- fasterize::fasterize(y, x, field = "ID")
  vals <- values(rast_poly)
  out <- raster::extract(x, seq_len(ncell(x))[!is.na(vals)], df = TRUE)
  out[["ID"]] <- vals[!is.na(vals)]
  out
}
library(rbenchmark)
benchmark(ex_values = raster::extract(x = srtm, y = zion, df = TRUE),
fex_values = fast_extract(srtm, zion), 
replications = 2)
#test replications elapsed relative user.self sys.self user.child sys.child
#1  ex_values            2   9.773   12.402     9.776        0          0         0
#2 fex_values            2   0.788    1.000     0.788        0          0         0


ex_values = raster::extract(x = srtm, y = zion, df = TRUE)
fex_values = fast_extract(srtm, zion)
identical(ex_values, fex_values)
#[1] TRUE

from geocompr.

Hey @mdsumner! Thanks for you kind words and sharing your knowledge.
I like the idea for fast_extract(), but it is probably too abstract for the book. It is also (I assume) more limited that raster::extract(), e.g. extracting by points or lines are not supported.

Yep, I am waiting for more consistent sf support in raster. Right now, crop expects sp object, while mask can also work on sf ones...

from geocompr.

fwiw this suggests it shouldn't matter any more: rspatial/raster#5

from geocompr.