If one has a weighted graph, would it make sense for the layout algorithm to account for the weights?
This is done in, for instance, the R package qgraph.
Perhaps I'm viewing this too simplistically, but I think that all that needs to be done is multiply by the weights at some point. For example, in the Spring layout, there is now
|
if adj_matrix[i, j] != zero(eltype(adj_matrix)) || adj_matrix[j, i] != zero(eltype(adj_matrix)) |
|
# F = d^2 / K - K^2 / d |
|
F_d = d / K - K^2 / d^2 |
|
else |
|
# Just repulsive |
|
# F = -K^2 / d^ |
|
F_d = -K^2 / d^2 |
|
end |
where, if I understand the code correctly, this line F_d = d / K - K^2 / d^2 could be changed to account for the weight of the edge. One option is to simply weigh the attraction by the (normalized) edge strength, which is I think what qgraph does (see https://github.com/cran/qgraph/blob/57d588ee05a554f9d1078914e05ce03662c610b2/src/layout_rcpp.cpp#L111).
Any thoughts? I'd be happy to open a draft PR for a single layout (e.g., spring).
If one has a weighted graph, would it make sense for the layout algorithm to account for the weights?
This is done in, for instance, the R package
qgraph.Perhaps I'm viewing this too simplistically, but I think that all that needs to be done is multiply by the weights at some point. For example, in the Spring layout, there is now
NetworkLayout.jl/src/spring.jl
Lines 95 to 102 in 6574223
where, if I understand the code correctly, this line
F_d = d / K - K^2 / d^2could be changed to account for the weight of the edge. One option is to simply weigh the attraction by the (normalized) edge strength, which is I think whatqgraphdoes (see https://github.com/cran/qgraph/blob/57d588ee05a554f9d1078914e05ce03662c610b2/src/layout_rcpp.cpp#L111).Any thoughts? I'd be happy to open a draft PR for a single layout (e.g., spring).