Nodes not appearing at their time during timeline playback

[solipsy]

Hi,

I have a strange problem with a dynamic graph in Gephi 0.8.1. beta. It's a dynamic social network of actors, constructed from IMDB data. Each actor first appears at some time, and this is properly written in the GEXF file as a node spell. The edges between actors - dates when they appeared together in a movie - are also annotated with spells.

When I enable the timeline, pull it to the start time and start ForceAtlas2, Gephi nicely filters out the nodes and edges. However, it doesn't keep doing so. Timeline plays, but new nodes and edges are not appearing. Strange thing is, if I stop and immediately restart ForceAtlas2 while timeline plays, nodes and edges appear. So if I want to see the network evolution, I have to keep pressing the "Start" button very fast. It's pretty comical, but it works. Is this a bug? Note that with smaller networks of the same type it works - nodes and edges appear at their scheduled time.

Another thing is, autoplay for node size tied to degree doesn't work. Nodes are same size at the end as they are on the beginning, even though their degrees are changing.

I'm attaching the GEXF file, if someone wants to take a look. Thanks for your help, much appreciated!

[seinecle]

Hi,

Just in case, could you test it with Gephi 0.8.2 beta (the latest release) to check if these issues persist?

Best,

Clement

[solipsy]

Hi,
Just in case, could you test it with Gephi 0.8.2 beta (the latest release) to check if these issues persist?

Hi,

I tried the it with 082b version and indeed the first problem is solved, that is, the nodes and edges are now appearing when they should.
The autoapply on node size, howewer, still doesn't work as expected. You can try it for yourself with the file attached to the first post. Or maybe there's an issue with the file itself?

Should I explicitly write node degree in an attribute and add spells to it for autoapply to work?

Thanks,

Marko

[seinecle]

A preliminary check is to make sure your graph is recognized as dynamic in the "import report" stage.

If it is, then in the statistics panel you should see new metrics available, for the dynamic case. There, you run the (dynamic) degree computation. If you did not do that your values for degree won't be dynamic - that's just going to be the degree overall.

Then before running the the timeline you have to apply the ranking on degree in the ranking panel, making sure you toggle on the very little "chain" icon just next to the apply button. This "chain" button activates the continuous computation on degrees while the timeline runs.

Only then can you run the timeline and see your node sizes or colors evolve with degree.

Even there, a last thing can cause your degree ranking not to work: do you rank your degree locally or globally? That's set up with the little funnel-looking icon still in the ranking panel (only appearing if you have a dynamic graph opened).

- Having the ranking local means that the extreme (lowest and highest) values considered for the ranking are those in the current time window. This usually leads to high variations. Use it when your frame of reference is the network as it is displayed in the current time frame.

- Having the ranking set to global means that the extreme values considered are those of the whole graph, across the whole time period. In this case, if you have just one single extremely high value at just one point in tie, such as a degree of 2,500 on a particular date, while the rest of the time degree fluctuate between 0 and 10, then moving the timeline won't show you anything evolving. Because as compared to this 2,500 value, all other values will remain desperately flat. Only when you will hit this 2,500 value will the node in question look briefly bigger, and that's it.
If you encounter this problem, and if your research question needs you to rank degrees globally, check for outliers in the higher range of values. This should be possible to exclude them from the computation by using the "range" slider in the ranking panel (chomping it on the right), but in my attempts that failed. So consider deleting this outlier for the purpose of this viz, if possible.

Hope this contributes to your research question!

Best,

Clement

[solipsy]

Hi,
and yes, I didn't know that about dynamic degree. I was calculating static degree and then wondered why it doesn't work in a dynamic graph. I simply didn't see it on the stat panel, and even if I saw it, I doubt that I'd know what it was for.

I propose a redesign and simplification of the UI. If a graph is dynamic, Gephi should always calculate static and dynamic degrees, and there should be only one option for that, not two. Maybe it should warn the user if the graph is too big that dynamic calculation will take too much time, and give an option to turn that off. What do you think?

best,

Marko

[seinecle]

Interesting!

To make your suggestions traceable, please open a ticket on Github:
https://github.com/gephi/gephi/issues?utf8=%E2%9C%93

We would all need more documentation, improvements to the UI and on other fronts too. Gephi is in continuous development, so this will eventually come. You are welcome to join and contribute!

Best,

Clement

[solipsy]

Thanks again! I opened a proposal on GitHub.

[solipsy]

Hi again,

I'm sorry to report that the first problem - network not refreshing dynamically with new nodes - seems to persist in the 0.8.2.b. I tried it with a now, bigger network that spans a longer time, and it's still there. I used Gephi64.exe and edited the conf file so that it uses 64bit JRE and more than 4GB of RAM.

You can see it in action here: http://www.youtube.com/watch?v=bZM5i-wdb3I. From time to time it seems to stop evolving, then it continues. That is where I have to press Stop/Run button on layout to refresh the network. Should I write a ticket, and if I do, should I attach the offending file?

The whole post is here: http://virostatiq.com/how-the-social-ne ... over-time/

Thanks!