http://www.medialab.sciences-po.fr/publ ... ndices.pdf
All the best
Statistics:Posted by jacomyma — 14 Mar 2013 10:49
]]>
The paper is very useful. Could it be possibly be posted on Arxiv to make it more visible? (Arxiv allows for later revisions of the paper).
Best,
Clement
Statistics:Posted by seinecle — 14 Mar 2013 10:39
]]>
many thanks for your work.
Alex.
Statistics:Posted by abrailas — 16 Feb 2013 06:17
]]>
It's not yet in prepublication (we don't even know if it will be accepted), so I put it there so that you can take a look:
ForceAtlas2 Paper (+appendices)
I think you'll find what you asked in this paper.
Thanks to the people here that gave me hints about their uses and understanding of this algorithm. It helped me to orientate the paper.
Statistics:Posted by jacomyma — 20 Sep 2011 06:32
]]>
Currently, I just have the following descriptors from the blog post about ForceAtlas2:
- Scaling: How much repulsion you want. More makes a more sparse graph.
Gravity: Attracts nodes to the center. Prevents islands from drifting away.
Dissuade Hubs: Distributes attraction along outbound edges. Hubs attract less and thus are pushed to the borders.
LinLog mode: Switch ForceAtlas’ model from lin-lin to lin-log (tribute to Andreas Noack). Makes clusters more tight.
Prevent Overlap: Use only when spatialized. Should not be used with “Approximate Repulsion”
Tolerance (speed): How much swinging you allow. Above 1 discouraged. Lower gives less speed and more precision.
Approximate Repulsion: Barnes Hut optimization: n² complexity to n.ln(n) ; allows larger graphs.
Approximation: Theta of the Barnes Hut optimization.
Edge Weight Influence: How much influence you give to the edges weight. 0 is “no influence” and 1 is “normal”.
Statistics:Posted by nullusadinfinitum — 20 Sep 2011 03:24
]]>
Go to Installed Plugins and remove ForceAtlas2, them update your Gephi and you'll have it
Statistics:Posted by mbastian — 07 Jun 2011 02:15
]]>
Networking problem in http://gephi.org/updates/thirdparty/0.8 ... atlas2.nbm
Statistics:Posted by elijah — 06 Jun 2011 23:54
]]>
Eduardo
Statistics:Posted by eduramiba — 06 Jun 2011 20:22
]]>
Unzip it, it's the folder of the Netbeans module.
I'll also push it as a branch in Launchpad (Mathieu Bastian told me how to do it).
Of course, feel free to contribute and improve it. Also, I think we might put it in Gephi's trunk. Any reports welcome !
Statistics:Posted by jacomyma — 05 Jun 2011 10:50
]]>
Please let us know when the source code will be released. I have already some requests for improvements - for example, the new version have some problems with graph streaming that the previous version doesn't have. I hope that, with the source code available, I could help with these improvements.
Cheers,
André
Statistics:Posted by panisson — 04 Jun 2011 15:38
]]>
Statistics:Posted by jacomyma — 31 May 2011 13:37
]]>
Cheers,
Martin
[1] http://bazaar.launchpad.net/~gephi.team ... ayout.java
[2]
http://bazaar.launchpad.net/~gephi.team ... ayout.java
jacomyma wrote:Hi everybody! I'm back and I am happy.
I updated the whole ForceAtlas code to make it "modern" and the result exceeds my expectations. Just for you (since you're interested in ForceAtlas) there is my current version of the ForceAtlas2 plug-in that I will officially publish soon. It has the same formulas and options (though with different names) with some simplifications, and features two main improvements:
- A better auto-stabilization with no settings needed, except a "sort of" speed called "tolerance to swinging"
- A Barnes Hut optimization that approximates repulsion, from n² to n.log(n) complexity -> far more performance.
Feel free to try it, and I will write the paper on the basis of this code.
Here is the NBM to install in Gephi's "plug-in" center:
[Download ForceAtlas2]
NB: NEEDS Gephi 0.8alpha at least !
Statistics:Posted by martin.pernollet — 30 May 2011 17:18
]]>
]]>
aurelienb wrote:Nice optimization !
BTW I cannot install it. I have " The Plugin Installer found problem timeout of loading ForceAtlas 2, Quality Layout[org.webatlas.forceatlas2/1.3] while install the following plugins: "... During the plugin install (after select plugin in Tool>Plugin >Downloaded >Install plugin
=> I had the same warning message during the installation?BTW I cannot install it. I have " The Plugin Installer found problem timeout of loading ForceAtlas 2, Quality Layout[org.webatlas.forceatlas2/1.3] while install the following plugins: "... During the plugin install (after select plugin in Tool>Plugin >Downloaded >Install plugin
Statistics:Posted by seinecle — 18 Apr 2011 03:48
]]>
I update my gephi 0.8, and when I would like to re install the plugin, gephi tells me the plugin was already installed. (And it works perfectly).
So there is probably a bug with plugin installer...
Statistics:Posted by aurelienb — 17 Apr 2011 11:02
]]>
Statistics:Posted by aurelienb — 14 Apr 2011 13:33
]]>
I updated the whole ForceAtlas code to make it "modern" and the result exceeds my expectations. Just for you (since you're interested in ForceAtlas) there is my current version of the ForceAtlas2 plug-in that I will officially publish soon. It has the same formulas and options (though with different names) with some simplifications, and features two main improvements:
- A better auto-stabilization with no settings needed, except a "sort of" speed called "tolerance to swinging"
- A Barnes Hut optimization that approximates repulsion, from n² to n.log(n) complexity -> far more performance.
Feel free to try it, and I will write the paper on the basis of this code.
Here is the NBM to install in Gephi's "plug-in" center:
[Download ForceAtlas2]
NB: NEEDS Gephi 0.8alpha at least !
Statistics:Posted by jacomyma — 13 Apr 2011 15:14
]]>
About settings:
- Attraction, repulsion and gravity do what we expect them to do, these are the main settings that impact the shape of the graph.
- Autostab, Max displacement and Speed, impact performance and the swinging of nodes.
I have no generic advice for the "shape" settings (except attraction distribution and adjust by size, I think of it). But I can show you a way to improve performance by settings. Try these settings, very spectacular for a 1000 nodes graph for example:
Maximum displacement: 100
Autostab Strength: 95
Autostab sensibility: 0.98
Speed: 5
In the end, if it swings too much, just slow the speed manually. The spectacular point is that it converges extremely fast to a balanced position. Autostab at its best...
Statistics:Posted by jacomyma — 08 Mar 2011 11:48
]]>
jacomyma wrote:Tell what you would develop, unanswered questions, what you dislike and so on...
Something I would require as a force atlas user is a set of advices for settings the algorithm parameters. Actually I would both like to know how you work them, and discuss a thing I noticed. Usually, the node size consideration + repulsion by degree do a great job. I found few constants for having a clean graph, but I noticed the following problem: if I apply the same graph structure with larger nodes, I have to increase the repulsion force to avoid small non connected nodes to get stucked in the middle of some big connected nodes.
I found that setting repulsion = 200 * maxNodeRadius generally works well on my dataset. It's a very experimental fact that probably rely on my data. Do you have a similar experience?
Regards,
Martin
Statistics:Posted by martin.pernollet — 28 Feb 2011 12:37
]]>
jacomyma wrote:III. Spatialization quality
Here we compare ForceAtlas to other layouts such as OpenOrd, Yifan Hu and Früchtermann Rheingold, on different graphs.
We do not test performance, but quality, meaning: interpretability.
III.1. Are close nodes more connected ?
III.2. Are close nodes structurally equivalent ?
III.3. Are close nodes in the same modularity clusters ?
I confirm ForceAtlas has a better readability that FR layout for a lot of small (50 nodes) graphs: clusters are grouped with sense, and there is really fewer edge intersection (if not none) than with FR layout.Here we compare ForceAtlas to other layouts such as OpenOrd, Yifan Hu and Früchtermann Rheingold, on different graphs.
We do not test performance, but quality, meaning: interpretability.
III.1. Are close nodes more connected ?
III.2. Are close nodes structurally equivalent ?
III.3. Are close nodes in the same modularity clusters ?
If you define a quality metric, then we can apply it to our graphs if that helps you comparing some layout. It won't be loosed time for us, since it will force us to dive into more rigorous selection of a layout.
Regards,
Martin
Statistics:Posted by martin.pernollet — 22 Feb 2011 13:02
]]>