How to have condition on SimpleWeightedDiGraph function?

Specific Domains Graphs
,
1

Hi Guys

According to the documentation the way SimpleWeightedDiGraph works is something like a g = SimpleWeightedDiGraph(heads, tails, weights). If so and if having a three list of requirements:

heads = [ 62, 17, 15, 22, 19,14,61, 21,51,35]
tails =     [62, 57, 29, 20, 19, 31,61,13,51,35]
weights = [309, 310, 307,308,302,316,321,301,302]

Is it possible when creating graph, directly exclude those edges that has same head and tails (eliminating loops) ?

And it seems that when we create g with graph package it starts at 1 regardless of the smallest node ID. For instance, in this simple example, if we call vertices(g) it return 1:62. This creates lots of single nodes that has degree of zero and makes my computations difficult when I want to work with node IDs that starts at 1000. I mean I have 1000 nodes that are empty and makes my graph very spars. When I call density on g it shows a very small number although my graph should be very dense. Is there any way to force SimpleWeightedDiGraph to not start at 1 but at the smallest node ID?

Thanks

2

Hi,
The best answer to both of your questions at the moment is to take care of it manually outside of the graph structure.
For self-edges, it is not hard to scan heads and tails and exclude pairs where both are identical:

new_heads = [h for (h, t, w) in zip(heads, tails, weights) if h != t]
new_tails = [t for (h, t, w) in zip(heads, tails, weights) if h != t]
new_weights = [w for (h, t, w) in zip(heads, tails, weights) if h != t]

Regarding vertex indices, the convention in the Graphs.jl ecosystem is that vertices should go from 1 to n. If you want to circumvent that, you can either

  • Translate the indices manually by keeping a dict outside of the graph
  • Use a more sophisticated structure for graphs with metadata, such as MetaGraphsNext.jl
1 Like
3

Thanks @gdalle for the info :slight_smile:

1 Like
4

Hi again @gdalle I just wanted to ask what do you suggest for translating indices?
The way I did it is in spreadsheet. But is there any direct way in julia to change IDs while keeping the relationship between nodes (how they form edges)?

5

For example, you have 15 nodes and suppose that we want to encode them.

node_id = union(heads, tails) |> unique! |> sort!
encode = Dict(node_id .=> 1:15)

replace!(heads, encode...)
replace!(tails, encode...)

and then just duplicate of @gdalle, we get below.

julia> new_heads = [h for (h, t, w) in zip(heads, tails, weights) if h != t]
5-element Vector{Int64}:
 4
 3
 8
 2
 7

julia> new_tails = [t for (h, t, w) in zip(heads, tails, weights) if h != t]
5-element Vector{Int64}:
 13
  9
  6
 10
  1

julia> new_weights = [w for (h, t, w) in zip(heads, tails, weights) if h != t]
5-element Vector{Int64}:
 310
 307
 308
 316
 301
1 Like
6

Thanks @rmsmsgood very much!

7

Is there any command like factorize in julia?
Python | Pandas Series.factorize() - GeeksforGeeks.

8

There are packages like CategoricalArrays.jl but that shouldn’t be needed here since all vertex indices and labels must be unique anyway

1 Like