Constraint rank of a node about d3-dag HOT 10 CLOSED

As of v0.4.3 the simplex layering support constraint rank. It's only tested very simply, so let me know if you encounter anything strange. In principle this could be added to the other layerings (with some caveats on topological), but especially adding the equality constraint makes things quite difficult, so I've deferred from implementing anything else.

Note, there are a lot of changes to API in 0.4, so see the upgrading guide if you see errors.

I'm going to close this for now as I think it's not required to add it to other layerings, but if you find a compelling reason as to why it should be, this can be reopened.

from d3-dag.

It's an interesting idea. Currently none of the methods support it, but you could always write your own "layering" operator that respected some notion like this. In principle though some of the methods could support something (especially simplex, as that constraint could just be added).

Do you have some idea as to what the api might look like? e.g. how you would specify a "rank" and what the contract might specify.

from d3-dag.

This would apply in time constrained graphs. Take this graph for example:

I'd like to have the ability to constraint node 2 and 8 to have the same rank as 21, for example. In DOT, you can specify the rank of a node by doing something like a [rank=1] and the algorithm lays out nodes with the same rank at the same depth.

API wise, I would just add a rank or depth (as rank may be confusing with the graph level property) property at the node level, just like you specify width and height. The rank would be a number which the user has to assign manually. In my case, I already know the exact rank of each of my nodes. My real problem is to optimize the layout horizontally to minimize edge crossings.

Does that make sense to you ?

from d3-dag.

Yeah, that seems reasonable. Given the current api, what I'm thinking is having something like a "rankAccessor" that returns a number or undefined. Any nodes that define rank as the same, must be put on the same level. Given the different layering operators, not all will support this, but it also seems reasonable to have them default to node.datum.rank.

In practice, since we're just looking for equality, these could really be anything, strings, booleans etc. In some sense, strings make more sense to me because of the equality and floating pointness of js numbers, but maybe number is more in line with dot. Thoughts?

from d3-dag.

rankAccessor is a good idea. I'm not sure what node.datum.rank is as I'm not aware of your entire API, but I'm assuming that any user defined properties are set into node.datum.

I think we shouldn't limit to equality checking. One day someone may want to say "I want node A to be one rank above node B". Numbers would be best, forget about floating points, people should use integers for this case as ranks are discrete, not continuous.

What do you think ?

from d3-dag.

This is what I've been looking for a while! Having a rankAccesor sounds great.
In the meantime, I was wondering how should my own "layering" operator look like? Are there any variables I must assign a value? Thanks in advance!

from d3-dag.

Sorry, I've been slow to actually implement this because I'm working (slowly) on the next major version that changes some conventions and adds typescript support. The actual change to implement this yourself shouldn't be too hard. You could mostly copy the simplex layering operator, and then modify it slightly to enforce the constraints imposed by rank accessor. I find my comments non-existent, and the framing of javascript-lp-solver unusual enough, that I can't give direct recommendation, but hopefully I'll have something in it soon.

from d3-dag.

@erikbrinkman please don't be sorry, none of us expects you to spend your time for free developing feature requests and we fully understand the perks of Open Source. I don't have the time to work on a PR as other priorities await, but I'll keep an eye on this issue.

from d3-dag.

@erikbrinkman it's fine. Just like @gempain said, nobody expects you to attend every feature request. Thanks for the advice! I'll try modifying the simplex layering operator.

from d3-dag.

@erikbrinkman thanks for the heads up and the quick implementation !

from d3-dag.