Morpher for calculating node neighborhoods based on spatial proximity

[luukvdmeer]

Is your feature request related to a problem? Please describe. A really nice morpher in tidygraph is to_local_neighborhood. For a given node, it limits the graph to the neighborhood of that node, based on a given order. E.g., order one gives all nodes that are directly connected to the root node, etc. When applied to spatial networks, this gives nice maps:

library(sfnetworks)
library(tidygraph)
#> 
#> Attaching package: 'tidygraph'
#> The following object is masked from 'package:stats':
#> 
#>     filter

net = as_sfnetwork(roxel, directed = FALSE)

plot(net)
plot(
  convert(net, to_local_neighborhood, 1, order = 10),
  col = "red",
  add = TRUE
)

^{Created on 2020-11-02 by the reprex package (v0.3.0)}

However, these neighborhoods are always computed based on the number of steps it takes to reach other nodes, not on the sum of edge weights it takes to reach other nodes. If that is possible, the neighborhoods actually are the same as isochrones (when using travel time as weights) or isodists (when using distance as edge weights). That would be a nice spatial morpher!

Describe the solution you'd like A to_spatial_neighborhood morpher that computes neighborhoods based on edge weights. Maybe igraph already has functionality to do this.

[luukvdmeer]

tidygraph::to_local_neighborhood wraps around igraph::make_ego_graph. An 'ego graph' is an induced subgraph of neighbors centered at node i within a given radius. This radius normally refers to the number of steps it takes to reach a node j from node i. In the Python package networkX, however, you can change this behavior and use edge weights instead. Hence, this allows calculating isochrones and isodists. See https://networkx.org/documentation/stable//reference/generated/networkx.generators.ego.ego_graph.html.

It seems unfortunately that igraph does not have this functionality. Seems like we have to implement this ourselves.

[agila5]

Hi! A (maybe naive) suggestion:

# packages
suppressPackageStartupMessages({
  library(sf)
  library(tidygraph)
  library(sfnetworks)
})

# data
roxel_sfn <- as_sfnetwork(roxel, directed = FALSE, length_as_weight = TRUE)

# Find nodes within 300m
roxel_sfn_small <- roxel_sfn %>% 
  filter(node_distance_from(1, weights = weight) <= 300)

# Plot
par(mar = rep(0, 4))
plot(roxel_sfn)
plot(roxel_sfn_small, col = "red", add = TRUE)

^{Created on 2020-12-06 by the reprex package (v0.3.0)}

This is not exactly the ideal approach since node_distance_from() calculates all the distances at the C level and then it filters at the R level. It would be better to filter the nodes at the C level (and then apply slice() or similar approach) but this is as far as I can go.

[luukvdmeer]

Amazing! I did not know node_distance_from accepted edge weights. I agree it could be optimized, but really like this as a first implementation, since it is also neat and clear. And also, if you want to do large isochrone/isodists calculations on large street networks, you'll probably be better of with the graphhopper/osrm bridges either way. No need to compete with that already ;)

[agila5]

Great! If you want sometimes next week I can add a PR with the new morpher.

[luukvdmeer]

Perfect! No time pressure ;) For the time being I will already include the workflow you presented in the new routing vignette.

[luukvdmeer]

The workflow is now presented in the routing vignette.

[luukvdmeer]

This is now part of the new release, thanks to @agila5

See this section in the morphers vignette for an example.