Add force-directed graph #25
Conversation
tofuness
changed the title
|
Exciting! Will review soon, thanks :) |
There was a problem hiding this comment.
Thanks @tofuness. This is really exciting!
I added some general comments here. And I wonder if you have any new tweaks to the jittery problem that we discussed.
|
|
||
| # Hacky Edge implementation | ||
| class Edge | ||
| constructor: (source, target) -> |
There was a problem hiding this comment.
Instead of a new class with source and target, I wonder if this can be just a Pair or Line
eg: var edge = new Line( source ).to( target ) ?
There was a problem hiding this comment.
The Edge implementation holds a reference to the source and target. The Line/Point implementation currently just copies the coordinates from source/target.
Two solutions I can currently think of:
- We change
Graph.addEdge(Edge edge)toGraph.addEdge(Point, Point)orGraph.addEdge([Point, Point]) - Let
Pairalso accept references toPoint.
There was a problem hiding this comment.
I see. Then I think either Graph.addEdge(P, P) or using an Edge class will make sense. Changing Pair to accept references seems a bit more dangerous :)
|
|
||
| @property 'p1', | ||
| get: -> | ||
| return { |
There was a problem hiding this comment.
Can be simplified as @target.clone()
| return new Edge(@source.clone(), @target.clone()) | ||
|
|
||
|
|
||
| class Vertex extends Vector |
There was a problem hiding this comment.
Can this be a Pair, or do you prefer the more semantic name displacement instead of p1?
| @height = height | ||
| @frames = [] | ||
|
|
||
| _fa: (x, k) -> |
There was a problem hiding this comment.
May want to move _fa, _fa, and _cool inside generate, since they are only used inside it.
| return @ | ||
|
|
||
|
|
||
| class ForceDirectedGraph extends Graph |
There was a problem hiding this comment.
What do you think if we simply make this as Graph, and then rename the generate() to forceDirected() ?
| class ForceDirectedGraph extends Graph | ||
| # Fruchterman & Reingold algorithm: http://citeseer.ist.psu.edu/viewdoc/download?doi=10.1.1.13.8444&rep=rep1&type=pdf | ||
| # Walshaw's algorithm: http://jgaa.info/accepted/2003/Walshaw2003.7.3.pdf | ||
|
|
There was a problem hiding this comment.
Thanks for reading all the papers! 👍
|
|
||
| v.displacement.add( | ||
| delta.$divide(deltaMag) | ||
| .multiply(@_fr(deltaMag, k)) |
There was a problem hiding this comment.
Would prefer moving this to a single line: $divide(...).multiply(...)
|
|
||
| if (v.x < 0) | ||
| v.x = 0 | ||
| if (v.x > @width) |
There was a problem hiding this comment.
use else if since v.x cannot be < 0 and > width.
Alternatively, use the Point's max and min functions: v.max( 0, 0 ).min( @width, @height)
| super() | ||
| @width = width | ||
| @height = height | ||
| @frames = [] |
| @@ -53,7 +53,8 @@ var coreFiles = coreElems.map(function(n) { return path.src.core+n+".coffee"; } | |||
|
|
|||
| var extendElems = [ | |||
| "Easing", "GridCascade", "ParticleEmitter", "ParticleField", "QuadTree", | |||
| "SamplePoints", "StripeBound", "UI", "Noise", "Delaunay", "Shaping" | |||
| "SamplePoints", "StripeBound", "UI", "Noise", "Delaunay", "Shaping", | |||
| "ForceDirectedGraph", "Vertex", "Edge" | |||
There was a problem hiding this comment.
Not sure if we should add Vertex and Edge as separate classes right now, unless they are useful enough on their own.
|
Sorry for keeping this PR stale 🙏 I've been thinking about the implementation and I think I need to revisit this algorithm. Fruchterman & Reingold (current implementation) runs on O(n^2) and is horribly slow. I lost my (half-working) implementation of this 😭 I've also looked into FADE (that builds on the Barnes-Hut simulation): I am not sure what I should do at the moment. Maybe try and take a stab at FADE? |
|
@tofuness Wow these papers are intense. I would suggest you pick an implementation since you probably know way more about this than I do. I appreciate your contribution, even if it's O(n^2). It's a start! |
Let's get this party started 🎊
demo/index.htmlfile and thegulpfile.jsto make this easier to review. Just rungulp extendand check out the demo to get started 😄 .Problems:
ForceDirectedGraphfile, I have multiple classes implemented.Graphto maybe extendPointSetEdgeto maybe extendPairEdgeimplementation is bad and looks like that because I wanted it to be compatible withForm.lines.Let me know your thoughts. Please take a look at this when you have time and sorry for not splitting it into multiple commits 💦 ..