create_LINESTRINGS_benchmark.R

# So this is just a test to see the fastests way to go from a dataframe with
# two coordinates per row (split into 4 columns - 2 for x and  2 for y) to a 
# linestring for each set of coordinates. 


# minimal example.
library(sfheaders)
library(wk)
library(sf)
library(tidyverse)
library(microbenchmark)

#data
rand <- function(n) sample(1:100, n, replace=T)
d <- tibble(x1=rand(1000),x2=rand(1000),y1=rand(1000),y2=rand(1000))

 
# original with purrr slow on large data
sf_purr_lines <- function(x){
  x %>%
  pmap_dfr(function(...){
    tibble(...)%>%
      mutate(g=st_sfc(st_linestring(matrix(c(.$x1, .$x2, .$y1, .$y2), 
                                           ncol=2), dim = 'XY')))%>%
      st_as_sf(sf_column_name='g')})
}

# sf_purr_lines(d) %>%
#  ggplot() + geom_sf()

# original but using rowwise() - better but still slow
sf_rowwise_lines <- function(x){
  x %>%
  rowwise() %>%
  mutate(g=st_sfc(st_linestring(matrix(c(x1, x2, y1, y2), ncol=2), dim = 'XY'))) %>%
  st_as_sf(sf_column_name='g')%>%
  ungroup()
}

# sf_rowwise_lines(d) %>%
#   ggplot() + geom_sf(colour='red')

# now with wk 
wk_lines <- function(x){
  df <- x %>%
  mutate(id=row_number())%>%
  pivot_longer(c(x1,x2,y1,y2),
               names_to = c(".value", "loc"),
               names_pattern = "(.)(.)")

  tibble(wk_linestring(xy(df$x,df$y),feature_id = df$id)) %>%
    st_as_sf()
}

# wk_lines(d)%>%
#   ggplot() + geom_sf(colour='blue')

#wk_filters
 wk_filters_lines <- function(x){
   df <- x %>%
     mutate(id=row_number())%>%
     pivot_longer(c(x1,x2,y1,y2),
                  names_to = c(".value", "loc"),
                  names_pattern = "(.)(.)")
   
   wk_handle(xy(df$x,df$y), 
             wk_linestring_filter(feature_id = df$id, 
             sfc_writer())) %>%
     st_as_sf()
   
 }
 # db <- tibble(x1=rand(100),x2=rand(100),y1=rand(100),y2=rand(100))
 # wk_filters_lines(d)%>%
 #      ggplot() + geom_sf(colour='pink')
 
# sfheaders solution 
sfheaders_lines <- function(x){
  x %>%
  mutate(id=row_number())%>%
  pivot_longer(c(x1,x2,y1,y2),
               names_to = c(".value", "loc"),
               names_pattern = "(.)(.)") %>%
  sf_linestring(., x='x', y='y', linestring_id = "id")
}

# sfheaders_lines(d) %>%
#   ggplot() + geom_sf(colour='green')


# let's benchmark it...
microbenchmark(
  sf_purr_lines(d),
  sf_rowwise_lines(d),
  wk_lines(d),
  wk_filters_lines(d),
  sfheaders_lines(d), times=10
)

d2 <- tibble(x1=rand(100),x2=rand(100),y1=rand(100),y2=rand(100))%>%
  sfheaders_lines()

p <- ggplot(sfheaders_lines(d)) + 
  # geom_sf(data=d2, 
  #         colour='black',alpha=0.7) +
  geom_sf(aes(colour=id), alpha=0.1) +
  scale_colour_viridis_c(option='rocket', guide='none') +
  theme_void()
p
# ggsave('exports/mako10k.png', p)