set.values assigns to all layers

[Jean-Romain]

set.values() assigns to all layers without being able to modify a specific layer

library(terra)
s <- rast(ncols=5, nrows=5, nlyrs=3)
set.values(s, 1:10, runif(10))
plot(s)

The only way I can find to touch a layer is

s[[2]][cells] <- value

Which involves a complex call stack and deeps copies. When touching multiple layers the code becomes extremely slow.

I think that providing a way to assign values to cells per layer would be useful.

Thanks

[rhijmans]

You can do

library(terra)
s <- rast(ncols=5, nrows=5, nlyrs=3)
set.values(s, 1:10, runif(10))

# replace an entire layer
s[[2]] = s[[3]] / s[[1]]
# replace cells
s[[1:2]][4] = 8

That seems to work OK. I suppose it becomes slow with large objects, as they may be read from disk, and written to temp files. In those situations, this kind of interactive use is less well supported. It would be possible to update existing files but that has its own risks; and I am wary of it. And if things work for a single layer, you can always operate on that layer, and re-insert it into the object. Perhaps an in place set.layer(x, i, y) (update layer i in x with y could help?

[rhijmans]

You can now do things like

library(terra)
s <- rast(ncols=5, nrows=5, nlyrs=3)
set.values(s, 1:10, runif(10), layer=2)

set.values(s, 11:20, cbind(1, runif(10)), layer=2:3)

[rhijmans]

A future refinement could be to not force all values (of the data sources changed) to memory by (optionally) allowing to update values on (temporary?) files, or else by creating temporary files where necessary.

[rhijmans]

It is now also possible to use a third index, k, to reference the layer when replacing or subsetting a SpatRaster with [ and [<-

library(terra)
s <- rast(ncols=5, nrows=5, nlyrs=3)
set.values(s, 1:10, 1:10)

s[3, 3, 3] <- 100
s[20, k=1:2] <- 18
s[cbind(4,4,3)] <- 50

s[3, 3, 3]