I am not sure if the travel-based graph is always what you want here.

if you consider what they are trying to capture with the graph in their paper, i'd argue it definitely IS what they want.

You may want contiguity that is just restricted in some way, which is when this comes in. But yeah, conceptually it is not the cleanest operation.

yeah, I'm just pretty skeptical of this approach. Contiguity graphs really come from research on things like states and counties where the entities are defined, a border is real, and a shared border means something explicit. Bastardizing a contiguity graph designed to represent 'neighborhoods' without considering what that means is definitely a bad idea IMO. Maybe the travel network isn't what you think you want, but then why does the highway create the disconnection in your graph? (it's because, presumably, you cant traverse across the highway so it creates a barrier between two units). But you have no idea if that's true... That's like assuming any passageway crossed by a raised L train is non-traversable, even though you can walk right under.

(so like, in urban space, a highway or some other barrier never creates a disconnection, it changes the travel impedance between two areas if you're required to travel around it--in which case it's better to measure that directly via a graph that accounts for distance). I guess what I'm saying is even if you 'want' a contiguity graph, which is already a weak abstraction for this use-case, then confounding it even more with sketchy assumptions is an bad idea. More realistically, you have barrier like a highway or water feature that forces you through a specific piece of the network (like an overpass or a bridge)

in which case, there are much better ways of measuring that. My recommendation here is to get people to think more critically about what their Graph actually represents (and my own 2 cents is the blunt removal of connections doesn't capture the break in social interaction potential they are actually trying to capture in that paper). I'd be much more inclined to include some sort of demo that illustrates the idea of editing the graph in some meaningful way rather than including a helper method. If you are going to do this, you should have a really strong grasp on what you're doing and why.

Uh oh!

Remove graph edges based on barriers #775

Description

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions