Parallel Processing with r5r

Question

I'm currently testing out the r5r package, which describes its processes as using parallel computation by default. Given the large number of origin-destination points I want to analyze quickly using its detailed_itineraries function, I wanted to test to see whether it could be sped up at all using any of the other parallelization tools in R.

I am trying using this code:

library(r5r)
library(sf)
library(tigris)
library(future.apply)
library(rJava)
library(tidyverse)

r5r_core <- setup_r5(data_path = path, verbose = FALSE)

tracts2 <- tracts(state = "PA", county = "Philadelphia", year=2019)%>%
  select(GEOID)%>%
  st_centroid()%>%
  st_transform("EPSG:4326")%>%
  arrange(GEOID)%>%
  rename(id = GEOID)%>%
  mutate(lon = unlist(map(geometry,1)),
         lat = unlist(map(geometry,2)))%>%
  st_set_geometry(NULL)%>%
  as.data.frame()

mode <- c("WALK", "TRANSIT")
max_walk_time <- 30 # minutes
departure_datetime <- as.POSIXct("14-06-2023 8:30:00",
                                 format = "%d-%m-%Y %H:%M:%S")

plan(multicore)

fn <- function(x, y){
  detailed_itineraries(r5r_core = r5r_core,
                       origins = x,
                       destinations = y,
                       mode = mode,
                       departure_datetime = departure_datetime,
                       max_walk_time = max_walk_time,
                       walk_speed = 4.5,
                       max_trip_duration = 60,
                       shortest_path = TRUE,
                       all_to_all = FALSE,
                       drop_geometry = TRUE,
                       progress= TRUE)
}

future_mapply(fn, tracts2, tracts2)

And am getting this error:

Error in assign_points_input(origins, "origins") : 
  'origins' must be either a 'data.frame' or a 'POINT sf'.

What is going wrong here? Alternatively, am I barking up the wrong tree trying to gain any speed this way?

Answer 1

{r5r} functions cannot be sped up using parallelization from R. We've tested it a few times already, and any parallelization we tried doing in R was less efficient than the current parallelization implemented in Java. To control the number of threads used when routing, please use the n_threads parameter.

PS: The error you're seeing results from incorrect future_mapply() usage. Under the hood, a data.frame is a list with some additional attributes and methods, so when you pass a data.frame to future_mapply() the function iterates to each one of its columns. Effectively, what you're doing with your code is passing tracts2's columns to detailed_itineraries(), not tracts2 itself.