[SOLVED] ForceAtlas2 prevents script returning

[jthrilly]

Hello everyone.

I have written a simple script which accepts a path to a graph file as a parameter, calculates modularity, colours the components based on this modularity, lays the graph out, and then writes back the file. Everything seems to work as expected.

My problem is that when I call the jar file from the command line, the script never returns me to the command prompt. This *only* happens when I attempt to use force atlas 2. The example documentation of Fruchterman-Reingold returns as expected, for example, and the script also returns when I don't do any layout at all. One strange observation is that if I open the gexf file, the layout seems to have completed as expected when using force atlas 2.

Here is the code I am using:

Code: Select all

        //Do layout
        ForceAtlas2 f2 = new ForceAtlas2Builder().buildLayout();
        f2.setGraphModel(graphModel);
        f2.setScalingRatio(2.0);
        f2.setGravity(7.0);
        //f2.setThreadsCount(12);
        f2.setLinLogMode(true);
        f2.setAdjustSizes(true);
        f2.initAlgo();
        for (int i = 0; i < 1 && f2.canAlgo(); i++) {
            f2.goAlgo();
        }


I have tried varying the number of iterations in the loop, but it has no effect.

------------------------------------------------------------------------------
SOLUTION:

In this case the solution was to make sure that endAlgo() is called after the for loop. The complete code therefore looks like this:

Code: Select all

//Do layout
        ForceAtlas2 f2 = new ForceAtlas2Builder().buildLayout();
        f2.setGraphModel(graphModel);
        f2.setScalingRatio(2.0);
        f2.setGravity(7.0);
        f2.setThreadsCount(12);
        f2.setLinLogMode(true);
        f2.setAdjustSizes(true);
        f2.initAlgo();
        for (int i = 0; i < 1000 && f2.canAlgo(); i++) {
            f2.goAlgo();
        }
       f2.endAlgo();


Because ForceAtlas2 is a continuous layout algorithm, it must somehow "stay alive" even after the iterations have been completed. This was not documented anywhere, so I think a note on the wiki (perhaps on this page: http://wiki.gephi.org/index.php/Toolkit_-_Layout_graph) about this specific algorithm would be a good idea.

[eduramiba]

Hi,
Some layouts such as force atlas 2 don't auto-stop even when it is stabilized. That means canAlgo will always return true, so you should set a maximum number of iterations in your for loop.

Edit: I see you put 1 iteration. I don't see anything wrong with the code. How are you loading your graph?

Eduardo

[jthrilly]

Thank you for your reply - makes perfect sense.

Is there a reason that forceatlas 2 does not stop even though it is stabalised? Is there something I can use to check if it has stabalised? The reason I ask is that the graphs will be of different sizes, and it is unclear to me how I should decide an appropriate number of iterations based solely on node/edge count.

[eduramiba]

Hi,
I guess it never stops because it is difficult to determine when the layout is done for any graph for sure.
Also I think Force Atlas 2 converges quickly at the beginning and slowly when it is more stable ir order to reach very good quality. You can see an example of the slow convergence here https://gephi.org/2011/forceatlas2-the- ... rew-layout

Did you manage to make that code work? If you execute that just 1 iteration it should return soon. Is the graph really big?
Does the layout work for that gexf file in Gephi 0.8.1-beta?

And if you don't want to manually set the number of iterations, you can execute the layout for an amount of time using AutoLayout http://wiki.gephi.org/index.php/Toolkit_-_Layout_graph

Eduardo

[jthrilly]

Edit: I see you put 1 iteration. I don't see anything wrong with the code. How are you loading your graph?

Yes, I did that when I was debugging to check it wasn't just happening because of too many iterations. The graph is loaded by:

Code: Select all

//Import file       
        Container container;
        try {
            File file = new File(args[0]);
            InputStream is = new FileInputStream(file);
            container = importController.importFile(file);
            container.getLoader().setEdgeDefault(EdgeDefault.UNDIRECTED);   //Force UNDIRECTED
        } catch (Exception ex) {
            ex.printStackTrace();
            return;
        }

Why would the way I load the graph affect ForceAtlas 2 but not other layout routines?

In response to your questions, yes the layout renders fine in standalone gephi - I am attempting to recreate the layout I get there. The networks are mostly small (<1000 nodes) with a mean of ~140.

I have considered using the automatic layout approach, but I don't wish to set a specific amount of time given the variation in the size of the networks. I really just want to understand why this approach isn't working.

EDIT: and no, with one iteration I am not sent back to the command prompt.

[eduramiba]

Oh, I think I know what is happening. It must be this bug https://github.com/gephi/gephi/issues/487.

It should be fixed with the latest toolkit. Make sure you are using it https://launchpad.net/gephi/toolkit/too ... .5-all.zip

Eduardo

[jthrilly]

I tried the link you suggested, but even though it says it is version 8.5 it unpacks as 8.1 and seems to contain the same gephi-toolkit.jar as I am already using. Either way, it does not solve my problem

Aren't the newer sources on GitHub? Can anyone try to reproduce my issue?

Thank you very much for helping me by the way!

[eduramiba]

I am going to check the file to see if it is the correct version.
Meanwhile you can try to add System.exit(0); at the end of your code as a workaround to fix it.

Eduardo

[jthrilly]

I figured out the problem. I didn't call endAlgo() after the for loop. It seems so obvious in retrospect, but it isn't documented anywhere that it is needed. I will update the OP and the issue thread on GitHub with this information.

Thank you Eduardo.

[eduramiba]

Right, call to endAlgo is missing in the toolkit examples. I will update them.
In the case of force atlas 2 it closes a thread pool.

Eduardo