#3791: r.stats.zonal with -r option
-------------------------+-------------------------
Reporter: frankie | Owner: grass-dev@…
Type: enhancement | Status: new
Priority: normal | Milestone: 7.4.5
Component: Raster | Version: 7.2.0
Keywords: | CPU: Unspecified
Platform: Unspecified |
-------------------------+-------------------------
When using a -r option to produce a reclassified output raster as result,
the r.reclass call at the end of the program could be *extremely*
inefficient. I tried with large maps many thousands of cats (derived by
r.clump) and I got very long running times.
Creating the rules files instead is very fast, even in those conditions.
In many cases, it would be sufficient giving it as an (optional) output
file and the r.reclass step could be avoided/postponed. BTW, undestanding
why the r.reclass module could take ages would be nice, too.
Replying to [ticket:3791 frankie]:
> When using a -r option to produce a reclassified output raster as
result, the r.reclass call at the end of the program could be *extremely*
inefficient. I tried with large maps many thousands of cats (derived by
r.clump) and I got very long running times.
That's why the default is not to create reclass maps.
>
> Creating the rules files instead is very fast, even in those conditions.
In many cases, it would be sufficient giving it as an (optional) output
file and the r.reclass step could be avoided/postponed. BTW, undestanding
why the r.reclass module could take ages would be nice, too.
The reason is that the underlying library function to assign category
labels `Rast_set_cat()` becomes terribly slow with an increasing number of
categories. The relevant parts of the raster lib would need to be
rewritten using a dynamic binary balanced search tree for a speed-up of
raster category handling.
I managed to work around as suggested, i.e. adding a rules= parameter to
save the reclassing rules file, then mimic r.category with simple egrep
clauses on the resulting rules set. At least it allows to have results in
decent time. I also noticed that while a lot of rules associate -nan they
are probably even processed in the r.reclass run. I agree that the right
fix would be modifying the library function, but in the meantime this
silly trick saved my butt
