Statement of Problem

I continue to require coffee at lunch time. If I was doing a Multi-Criteria Decision Analysis I’d throw in some requirements:

Be able to get to the coffee and back within a 30-60 minute lunch.
Be reasonably priced.
Taste good.
Preference for independent coffee shops. Cost/bucks &c. might be acting legally when they minimise their tax bill, but I can take a stand on the ethics of it by preferring local.
Getting a reasonable number of steps in is a nice-to-have. If it’s not raining.

Previous Solutions

I looked at this with bits of Python towards the start of this blog, and there’s been some developments in R packages for spatial work, interacting with the OpenStreetMap API, and network theory/graph theory.

Currently they’re not brilliant at talking to each other! But for these purposes it’s not terrible in computing time to load the data twice into different libraries to look at it from different angles - once as geography and once as a network connected by “can walk from A-B”.

Today’s Solution

Normally I put my library calls in the hidden bit at the top, but the serious work is being done by the libraries, so they’re going in the main post today:

library(tidyverse)
library(sf)
library(osmdata)
library(osmar)
library(tidygraph)
library(tmap)

First, I grabbed a sensible bounding box and all the cafes inside that box:



bbox <- c(-1.5853357315063479,53.78906626347749,-1.513667106628418,53.81142134499526)

cafe <- opq(bbox = bbox) %>%
  add_osm_feature(key = 'amenity', value="cafe") %>%
  osmdata_sf()

tmap_mode("view")
tm_shape(cafe$osm_points) + tm_dots()

So far so good, just need to trim this down to what can be feasibly reached in a lunch break.


# Grab footpaths
opq(bbox=bbox) %>% 
  add_osm_feature(key="highway") %>%
  osmdata_xml(here::here("static", "data", "OSM", "footpaths.osm"))

# This step is computationally expensive and I didn't want to do it every time something lower down went wrong.
expensive_parse <- xmlParse(here::here("static", "data", "OSM", "footpaths.osm")) %>%
  as_osmar() 

footpaths <- expensive_parse %>%
  as_igraph() %>% 
  as_tbl_graph() %>% # Footpaths are now in graph representation
  activate(nodes) %>%
  arrange(name!="5506378727") %>% # Manual work to make a node near the enterance to work the first node
  mutate(dist = bfs_dist(root=1)) %>% # Every node labelled with its footpath distance
  as_tibble() # We no longer need the representation as a graph.

# We do need the coordinates of each node to find which paths are near which coffee shops.
nodes <- expensive_parse$nodes$attrs %>% 
  st_as_sf(coords = c("lon", "lat")) %>%
  select(id)

# We get a tibble of coordinates and the shortest walking path distance to them.
path_distance <- nodes %>%
  mutate(name = as.character(id)) %>%
  select(-id) %>%
  left_join(footpaths) %>%
  filter(!is.na(dist)) %>%
  st_set_crs(4326)

As far as I can tell, st_nearest_feature doesn’t like me asking “for every coffee shop what’s the nearest footpath point”, so I’ve resorted to a map_dbl.


cafe_distance <- function(index){
  
  st_nearest_feature( cafe$osm_points[index,] %>% st_transform(27700)
                      , path_distance %>% st_transform(27700))  %>%
    first() %>%
    slice(path_distance, .) %>%
    st_drop_geometry() %>%
    select(dist) %>%
    dplyr::pull()
}

cafes <- nrow(cafe$osm_points)

cafe <- cafe$osm_points %>%
  mutate(dist = map_dbl(seq_len(cafes), cafe_distance))

Finally, plot it!

cafe %>%
  select(name, everything()) %>% # Makes name the first column, makes name the mouseover text
  tm_shape() + tm_dots( col="dist", palette = "viridis")

Next Steps

If I could eliminate the little bits of manual work I put into this, I could roll it into a shiny app, or generate coffee maps for arbitary points.

Open Coffee Map Part 4

Statement of Problem

Previous Solutions

Today’s Solution

Next Steps