Categorical `SpatRaster` layers break problem creation

[Martin-Jung]

Categorical `SpatRaster` layers break problem creation

[jeffreyhanson]

Ok sounds great - thanks @ricschuster and @Martin-Jung! I'll update prioritizr to throw errors if a user supplies a categorical raster.

[jeffreyhanson]

Hey,

Happy new years! Thank you very much for raising this issue! I imagine this could be confusing for a new user that's familiar with categorical rasters (honestly, I had no idea categorical rasters existed). I think the best option might be to either (1) convert them automatically to numeric rasters (using as.numeric()), or (2) throw a clear error saying that categorical rasters are not supported and tell the user to convert them manually.

Can you think of any issues/situations with (1) (i.e., using as.numeric()), where prioritizr might automatically convert a categorical raster to a numeric raster and, as a consequence, produce an output that doesn't align with the users intent? For example, imagine if a user has a categorical feature raster wherein different pixel values/classes correspond to different ecosystem types, then each pixel value is a different feature. In this case, I imagine using as.numeric() automatically in prioritizr would result in prioritizr incorrectly proccessing the data. This is because using as.numeric() would produce a single integer layer, when the user actually needed multiple binary layers so each ecosystem type is represented as a different feature. Is that right? Or maybe I'm misunderstanding something?

[ricschuster]

I personally would go with (2) and throw a clear error. Option (1) makes assumptions that might not be correct and I think it would be better if users need to think about what to do rather than the package assuming what a user might want.

What do you think?

[Martin-Jung]

Conversion factor to numeric 1):
Mhh, there could be potentially issues in this if a user has for some reason differently labelled categories that do not align with the numeric values. In this case numeric values are assigned based on the factor level order. For example:

a <- c("1", "3", "4") |> as.factor()
as.numeric(a)

So in that case I agree with Richard and think option 2) (assert_that call checking whether categorical rasters are supplied throughout) would be preferable.