Slow for `getMinBBox()`

[harryprince]

I am using getMinBBox for get implement st_orientedevelope function like this, and I find it is slower than python counterpart, does any idea for the solution?

p =  "POLYGON((116.4821384 39.9836878,116.4827035 39.9841856,116.4827233 39.9841873,116.4827841 39.9841542,116.4830736 39.9839845,116.4831764 39.9839125,116.4831793 39.9838734,116.4830778 39.9837773,116.4820606 39.9828711,116.4788117 39.9799144,116.4787834 39.9798948,116.478763 39.9798933,116.4784785 39.9800747,116.4783037 39.980195,116.4783008 39.9802338,116.4821384 39.9836878))"

microbenchmark::microbenchmark(
 t1 <-  p %>%
  sf::st_as_sfc() %>%
  sf::st_coordinates() %>%
    data.frame() %>%
  shotGroups::getMinBBox() %>% with(pts) %>% 
  data.frame() %>%
  sf::st_as_sf(coords=c("x","y")) %>% 
  summarise(do_union=F) %>%
  sf::st_cast("POLYGON")
  ,

 t2 <-  reticulate::py$shapely$wkt$loads(p)$minimum_rotated_rectangle$wkt %>%
   sf::st_as_sfc() 
  )

lwgeom::st_astext(t1$geometry) == lwgeom::st_astext(t2)

method	min	lq	mean	median	uq	max	neval	cld
t1	5.594826	5.878546	7.376063	6.106935	6.809007	16.335442	100	b
t2	3.052761	3.268867	3.901166	3.396058	3.706425	8.500606	100	a

[dwoll]

Thanks for your interest in shotGroups! It could be that getMinBBox() can be made faster. However, you can already save time by avoiding conversion between matrix and data frame. As I don't have a Python installation, I only compare two versions using getMinBBox() I don't know much about sf, the second function returns a POLYGON, not a "Simple feature collection".

p <- "POLYGON((116.4821384 39.9836878,116.4827035 39.9841856,116.4827233 39.9841873,116.4827841 39.9841542,116.4830736 39.9839845,116.4831764 39.9839125,116.4831793 39.9838734,116.4830778 39.9837773,116.4820606 39.9828711,116.4788117 39.9799144,116.4787834 39.9798948,116.478763 39.9798933,116.4784785 39.9800747,116.4783037 39.980195,116.4783008 39.9802338,116.4821384 39.9836878))"

do_shotgroups1 <- function(x) {
    x %>%
        sf::st_as_sfc() %>%
        sf::st_coordinates() %>%
        data.frame() %>%
        shotGroups::getMinBBox() %>%
        with(pts) %>% 
        data.frame() %>%
        sf::st_as_sf(coords=c("x", "y")) %>% 
        summarise(do_union=FALSE) %>%
        sf::st_cast("POLYGON")
}

do_shotgroups2 <- function(x) {
    pts <- x %>%
        sf::st_as_sfc() %>%
        sf::st_coordinates()

    bb <- shotGroups::getMinBBox(pts[ , 1:2])["pts"]
    bb[["pts"]] <- rbind(bb[["pts"]], bb[["pts"]][1, ])
    sf::st_polygon(bb)
}

microbenchmark::microbenchmark(t1 <- do_shotgroups1(p),
                               t2 <- do_shotgroups2(p))

expr	min	lq	mean	median	uq	max	neval	cld
t1 <- do_shotgroups1(p)	4.9574	5.22285	6.704090	5.51855	7.70995	22.5991	100	b
t2 <- do_shotgroups2(p)	1.1606	1.23370	1.522725	1.31400	1.48030	5.1822	100	a

[harryprince]

thanks, it seems the R version is faster than the Python counterpart.