Node placement problem

[d3]

Node placement problem

[cosmin-novac]

+1

I'm interested in a solution to this problem as well. I've found some ideas here, but haven't found a no concrete solution yet.

[cirofdo]

I came with a solution that I'm still working on, but it is a specific to the problem you have.

The idea is to change the "class" attribute to all nodes, so you can manually change its position.

For example:

• Links class: .attr("class", function(d) { return "link " + d.source.name + " " + d.target.name; })
• Nodes class: .attr("class", function(d) { return "node " + d.name; })

Now you can change the position on the .css.

PS: I'm really new to html, css and js programming (less than 4 months). If this is not recommended I'll be pleased to know.

[mbostock]

This library performs relaxation in passes of alternating directions, from right-to-left and then left-to-right. When relaxing from left-to-right, it moves each node’s following nodes (the targets of the node’s source links) towards the node’s center; for example, it tries to center B around the center of A. Likewise when relaxing from right-to-left, it moves each node’s preceding nodes (the sources of the node’s target links) towards the node’s center; for example, it tries to center B around the weighted average centers of C, G and D.

In this example, the primary problem is the left-to-right pass, where it moves the center of B up towards the center of A. (There’s a similar but lesser problem moving the center of B up towards the center of C as part of the right-to-left pass, but it’s mitigated because the weighted average center of C, G and D is used.)

A better heuristic would probably try to contain B within the span of A on the left-to-right pass (and similarly for the right-to-left pass), but simply changing to that heuristic instead of the move-to-center heuristic does not work well because it greatly increases the overlap between links. Currently, there is no penalty when a link passes through a node, such as the link from B to G that passes through C.

Fixing this problem in general requires a more thorough redesign of how the layout is optimized. It might also require the insertion of dummy nodes, so that when a link spans more than one topological layer (such as B→G), it can be routed through a dummy node to avoid overlap with other nodes in the intermediate layers (such as B→dummy→G, where the dummy then avoids collision with C).

So, the short summary of this issue is “the generated layout could be better” and the answer is “yes, but it’s hard”, and I may get around to working on it sometime in the future but I don’t have a concrete plan for doing so in the short term, and so I’m going to close this issue. However, contributions for better layout heuristics and algorithms are welcome!