Inefficiency

[mtennekes]

Hi @agila5 @luukvdmeer @Robinlovelace

Before submitting tmap.networks to CRAN, I'd like make the preprocessing step more efficient.

Could you help me a bit?

• In tmapGetShapeMeta1() the data variable names from the spatial objects are collected. I didn't now how to do this directly (as fast as possbile). See https://github.com/r-tmap/tmap.networks/blob/main/R/tmapShape_sfnetworks.R#L9C1-L37C2.
• In tmapGetShapeMeta2() the factor levels are determined (which will be used for faceting). See https://github.com/r-tmap/tmap.networks/blob/main/R/tmapShape_sfnetworks.R#L69-L86 Same question: can this be improved, e.g. without having to call to_spatial_explicit?

[Robinlovelace]

Quick answer: I'm not sure sorry.

[agila5]

Hi @mtennekes! I started checking the code that you provided, but I have a question. Why do you need to explicitly call to_spatial_explicit() when retrieving the data variable names? I don't see the link between "explicit edges" and "variable names". For example:

library(igraph)
library(tidygraph)
library(sfnetworks)

# Create sfn which is spatially implicit (i.e. no LINESTRING geometry)
sfn <- roxel |> 
  as_sfnetwork() |> 
  convert(to_spatial_implicit)

# get_names still works even when the underlying network is "implicit"
tmap.networks:::get_names(sfn, "nodes")
#> [1] ".tidygraph_node_index"
tmap.networks:::get_names(sfn, "edges")
#> [1] "name"                  "type"                  ".tidygraph_edge_index"

^{Created on 2025-05-06 with reprex v2.1.1.9000}

[mtennekes]

Good point! I've improved tmapGetShapeMeta1

How about tmapGetShapeMeta2? The purpose is to loop through all node-attribute data columns and all edge-attribute data columns and apply get_fact_levels_na, an internal function.

[agila5]

I guess the following might be slightly more efficient but probably not by much. The main benefit is that you don't need the sf::st_as_sf calls and all the other preliminary steps, just use igraph code:

library(igraph)
library(tidygraph)
library(sfnetworks)

# Create sfn which is spatially implicit (i.e. no LINESTRING geometry)
sfn <- roxel |> 
  as_sfnetwork() |> 
  convert(to_spatial_implicit)

o = list(drop.NA.facets = TRUE, facet.max = 9)

lapply(edge_attr(sfn), tmap:::get_fact_levels_na, o = o)
#> $name
#> NULL
#> 
#> $type
#> [1] "cycleway"     "footway"      "path"         "residential"  "secondary"   
#> [6] "service"      "track"        "unclassified"
#> attr(,"showNA")
#> [1] FALSE
#> 
#> $.tidygraph_edge_index
#> NULL
lapply(vertex_attr(sfn), tmap:::get_fact_levels_na, o = o)
#> $geometry
#> NULL
#> 
#> $.tidygraph_node_index
#> NULL

^{Created on 2025-05-06 with reprex v2.1.1.9000}

[mtennekes]

Thx @agila5 I see that to_spatial_implicit is not in the CRAN version of sfnetworks? When do you expect to update it? @luukvandermeer?

[agila5]

I'm not sure but, in any case, that's not super important for the examples that I presented. I just wanted to show that the code works even when the edges are implicit. Everything should work the same even with explicit edges.