[00:00.000 --> 00:22.000]  Okay, welcome everyone, as you can see from Adam, typical notebooks are a very important
[00:22.000 --> 00:32.000]  tool in each data scientist, but using graphs, refer to notebook as a challenge, for instance
[00:32.000 --> 00:33.000]  visualization.
[00:33.000 --> 00:41.000]  And so Bjornville talked about, I pie Zygma today, which is a tool to use ZygmaJas as
[00:41.000 --> 00:43.000]  a component in a Jupyter notebook.
[00:43.000 --> 00:47.000]  So I'm really looking forward to that and without further ado.
[00:47.000 --> 00:53.000]  Where I'm from this time, I'm actually not Guillaume, I'm sick.
[00:53.000 --> 01:01.000]  So I apologize in advance, because I'm not the creator of the tool and so I will do as
[01:01.000 --> 01:08.000]  much as I can to present it, but Guillaume can answer that by email or by Twitter or any
[01:08.000 --> 01:13.000]  other means if you have more questions than what I can actually answer myself.
[01:13.000 --> 01:21.000]  And so I will just start by a brief remember of why we sometimes want to use graphs and
[01:21.000 --> 01:28.000]  actually visualize them and not only do statistics on notebooks and actually visualize graphs.
[01:28.000 --> 01:30.000]  And so why do we do visual network analysis?
[01:30.000 --> 01:37.000]  It actually goes back very old to 1736 and the Bridges of Collector, which is a classical
[01:37.000 --> 01:44.000]  mathematical problem that was solved thanks to visualizing the graph that it was showing.
[01:44.000 --> 01:53.000]  Later on in France, Moreno did a social graph where he tried to visualize how a connective
[01:53.000 --> 01:56.000]  where students in a classroom.
[01:56.000 --> 02:06.000]  And recently, thanks to the community assisting tools, we can do those kind of visualizations
[02:06.000 --> 02:15.000]  but with massive graphs and we can try to do a computed processing to try and automatically
[02:15.000 --> 02:24.000]  specialize nodes on the map and on the plane and also identify clusters within it.
[02:24.000 --> 02:32.000]  So that brings a different mean to actually analyze graphs and actually visualize this
[02:32.000 --> 02:34.000]  helps a lot understanding.
[02:34.000 --> 02:41.000]  And we are coming from the field of social sciences and we use a lot of graphs to interpret
[02:41.000 --> 02:44.000]  social issues in general.
[02:44.000 --> 02:47.000]  And we use them actually as maps.
[02:47.000 --> 02:51.000]  So it's not maps in which coordinates make sense.
[02:51.000 --> 02:53.000]  X and Y don't mean anything.
[02:53.000 --> 02:55.000]  You can just take the map.
[02:55.000 --> 03:03.000]  But basically what you see on the plane indicates information on the, I mean the localization
[03:03.000 --> 03:07.000]  of each node that makes a sense compared to the other nodes.
[03:07.000 --> 03:10.000]  So, but I guess most of them are not.
[03:10.000 --> 03:16.000]  That's another example of a map that was made a long time ago.
[03:16.000 --> 03:27.000]  So to do that, there has been over the past years a lot of tools that have been developed
[03:27.000 --> 03:29.000]  including the first desktop ones.
[03:29.000 --> 03:38.000]  So this tool is the direct heritage of this long lineage which started with Gezi.
[03:38.000 --> 03:44.000]  I believe later today there will be a presentation of Gezi version 1 which finally will go out
[03:44.000 --> 03:48.000]  soon after so many versions already.
[03:48.000 --> 03:50.000]  So you probably already know Gezi.
[03:50.000 --> 03:58.000]  But recently we could switch from the actual desktop analysis to actual web representations
[03:58.000 --> 04:01.000]  thanks to a variety of libraries.
[04:01.000 --> 04:04.000]  D3.js proposes to do it.
[04:04.000 --> 04:07.000]  But there's also a site escape and a bunch of others.
[04:07.000 --> 04:11.000]  But so our community works with Simba.
[04:11.000 --> 04:17.000]  And Simba has been developed by people who are actually close to the people of Gezi.
[04:17.000 --> 04:19.000]  I don't think Alexi is here today.
[04:19.000 --> 04:24.000]  But Alexi Jacomi is the small brother of Mathieu Jacomi who speaks about Gezi.
[04:24.000 --> 04:31.000]  He's the one who invented Simba and Guillaume is the co-host of Simba with Alexi.
[04:31.000 --> 04:33.000]  So please take a look at Simba.
[04:33.000 --> 04:39.000]  I will put the slides to the conference and you will find links to all the tools around.
[04:39.000 --> 04:46.000]  And then thanks to Simba, we could build a lot of Gezi-like tools but for the web.
[04:46.000 --> 04:51.000]  So that we could do all those interactions that we do directly in a web page.
[04:51.000 --> 04:57.000]  There's been a long history at Miguelab and around of trying to build such tools.
[04:57.000 --> 04:59.000]  Minivan was one of them.
[04:59.000 --> 05:08.000]  There's also Nancy which is a very small, very publishing-oriented way of displaying a graph
[05:08.000 --> 05:16.000]  with very few options so that you can just put your Gezi-like search file or GraphML file
[05:16.000 --> 05:21.000]  and very easily do what you do on your Gezi.
[05:21.000 --> 05:25.000]  Retina is the one developed by people at Westware right now and is very rich,
[05:25.000 --> 05:27.000]  proposing a lot of features.
[05:27.000 --> 05:35.000]  And soon I think Mathieu and Mathieu will talk about it briefly also in the later talk about Gezi version 1.
[05:35.000 --> 05:41.000]  There's a Gezi-like version that's currently being developed and that should come in soon.
[05:41.000 --> 05:45.000]  Which brings me to all of those tools are very nice.
[05:45.000 --> 05:51.000]  We have all those that are interactive and you can visualize, explore, publish,
[05:51.000 --> 05:58.000]  manipulate all those graphs but they all require pre-processed graphs.
[05:58.000 --> 06:04.000]  You cannot just work with your graph while you're visualizing it.
[06:04.000 --> 06:11.000]  You have to pre-code in your file, usually JSON or JXF or GraphML,
[06:11.000 --> 06:14.000]  then you load it into the tool and then you can explore it.
[06:14.000 --> 06:17.000]  But we would like to be able to do that at the same time.
[06:17.000 --> 06:21.000]  And so that's where the idea of I by Sigma came from,
[06:21.000 --> 06:29.000]  to try and put within Jupyter a notebook, a widget that would display the graph using SigmaJS.
[06:29.000 --> 06:34.000]  So it's really easy to install as long as you have Jupyter.
[06:34.000 --> 06:39.000]  You usually need a tool to work with graphs under Python.
[06:39.000 --> 06:43.000]  There's two main ones that you might know already about,
[06:43.000 --> 06:53.000]  I by Sigma is built to handle both formats of graphs from both networks and I by Sigma.
[06:53.000 --> 06:56.000]  And so you just install I by Sigma in addition,
[06:56.000 --> 06:59.000]  and then I just switch to the brief demo.
[06:59.000 --> 07:02.000]  Maybe at the seat.
[07:02.000 --> 07:07.000]  So we'll do two small explorations of graphs.
[07:07.000 --> 07:11.000]  There's the first one that we're working on right now,
[07:11.000 --> 07:16.000]  which is on the, what I call the open source,
[07:16.000 --> 07:21.000]  I mean, actually larger than that, open access, open world.
[07:21.000 --> 07:28.000]  It's like for them, but just in France and the French communities working on that.
[07:28.000 --> 07:36.000]  And so we built this network of websites, links together of those French communities of free software.
[07:36.000 --> 07:38.000]  And let's take a look at it as well.
[07:38.000 --> 07:41.000]  So first, I will import the projects.
[07:41.000 --> 07:44.000]  Then I'm reading the graph that I built already.
[07:44.000 --> 07:46.000]  So that's all this first example.
[07:46.000 --> 07:52.000]  So here we have a graph with 621 nodes and 7000 edges.
[07:52.000 --> 07:54.000]  Let's look at the node.
[07:54.000 --> 07:58.000]  So the first node, I don't see that information.
[07:58.000 --> 07:59.000]  It's April.org.
[07:59.000 --> 08:02.000]  I don't know if the French people are in the room,
[08:02.000 --> 08:07.000]  but people should know that April is from France.
[08:07.000 --> 08:14.000]  It's the main NGO in France about open source and just the graph.
[08:14.000 --> 08:16.000]  So we have this whole page.
[08:16.000 --> 08:19.000]  That's all the data that was collected while making the graph.
[08:19.000 --> 08:25.000]  And then let's try to just visualize it by just loading i5 sigma,
[08:25.000 --> 08:28.000]  importing sigma and applying it to the graph.
[08:28.000 --> 08:37.000]  Here, just by the widget with the graph, which is randomly specialized.
[08:37.000 --> 08:40.000]  We have metadata information.
[08:46.000 --> 08:49.000]  So we can run for FATAS on it.
[08:49.000 --> 08:51.000]  So very easy.
[08:51.000 --> 08:54.000]  You see your specialized graph.
[08:54.000 --> 09:01.000]  Just a few seconds, and then we can also apply some...
[09:01.000 --> 09:05.000]  The graph is too dense for that.
[09:05.000 --> 09:07.000]  And suddenly no effect.
[09:07.000 --> 09:09.000]  So yeah, but right now it's just a graph
[09:09.000 --> 09:11.000]  and we don't have much information.
[09:11.000 --> 09:13.000]  It's very complicated readable.
[09:13.000 --> 09:18.000]  So let's go down and try to add a few other options to the sigma code.
[09:18.000 --> 09:21.000]  So we can set the outside standards.
[09:21.000 --> 09:23.000]  Let's use the number of pages for.
[09:23.000 --> 09:27.000]  So here I can see that for this graph use,
[09:27.000 --> 09:31.000]  we put a lot of pages on some specific websites.
[09:31.000 --> 09:36.000]  Let's put a little bit more and try to adjust the sizes of the nodes.
[09:36.000 --> 09:39.000]  So we can adjust the range of the values for instance.
[09:39.000 --> 09:42.000]  Here it's really readable.
[09:45.000 --> 09:47.000]  Okay, so we got sizes.
[09:47.000 --> 09:49.000]  Let's add some colors.
[09:49.000 --> 09:55.000]  So iBuySigma proposes some internal metrics that you can compute on the fly.
[09:55.000 --> 10:00.000]  So for instance, as a result, it generates clusters.
[10:00.000 --> 10:03.000]  And we will apply colors to the map.
[10:03.000 --> 10:07.000]  I mean we will apply those color communities as colors.
[10:07.000 --> 10:10.000]  So here we get a set graph of colors.
[10:10.000 --> 10:14.000]  Let's see that there are a lot of communities.
[10:14.000 --> 10:19.000]  As knowing this network and knowing this community,
[10:19.000 --> 10:23.000]  I can tell you that basically what this is.
[10:28.000 --> 10:35.000]  Here we got the open data, open command community.
[10:35.000 --> 10:41.000]  Here we got uphill and basically the NGOs working on the open source.
[10:41.000 --> 10:47.000]  Here we got GIL and it's mostly a lot of softwares.
[10:47.000 --> 10:51.000]  Fedora and all the Linux distributions.
[10:51.000 --> 10:54.000]  Here we got FFDL, MaproductionDenets,
[10:54.000 --> 10:59.000]  and all those activists working with the open internet.
[10:59.000 --> 11:03.000]  And I guess here is more the...
[11:03.000 --> 11:06.000]  Oh, it's also a mobilization.
[11:06.000 --> 11:11.000]  It's a form of formigated old form of food.
[11:11.000 --> 11:14.000]  I'll just speak a little bit.
[11:14.000 --> 11:19.000]  Okay, so now that we got this, let's try to make it a little bit nicer.
[11:19.000 --> 11:25.000]  We can add, for instance, some border colors.
[11:25.000 --> 11:31.000]  So it just proposes to see a stronger border of colors.
[11:31.000 --> 11:34.000]  Graphs are a little bit sexier.
[11:34.000 --> 11:39.000]  We can also try to do like Gephi, like curled edges.
[11:39.000 --> 11:41.000]  All of those are in options.
[11:41.000 --> 11:45.000]  I guess I'll show you briefly later on a list of the different options.
[11:45.000 --> 11:49.000]  Here we also put the recursive font to the level.
[11:49.000 --> 11:52.000]  So basically you can do a lot of things.
[11:52.000 --> 11:55.000]  But all of that so far is mostly like Gephi.
[11:55.000 --> 11:58.000]  There's no real new thing.
[11:58.000 --> 12:03.000]  But here's something that actually proposes something else.
[12:03.000 --> 12:06.000]  So right now we can see one graph.
[12:06.000 --> 12:10.000]  But let's try and see multiple ones.
[12:10.000 --> 12:14.000]  So I buy similar properties, what we call a similar grid.
[12:14.000 --> 12:19.000]  And so I will put the same graph, but it will trickle out.
[12:19.000 --> 12:24.000]  And those will be common options that I set for all versions of the graph.
[12:24.000 --> 12:28.000]  But then, within the grid, I will add three different versions of the graph
[12:28.000 --> 12:32.000]  using different metrics for the size of the load.
[12:32.000 --> 12:34.000]  So here's one on the left one.
[12:34.000 --> 12:37.000]  And we see it's on the middle of the degree.
[12:37.000 --> 12:41.000]  And the right one on the bottom.
[12:41.000 --> 12:46.000]  Now I'm going to add this.
[12:46.000 --> 12:53.000]  So here are the three graphs, which are all synchronized.
[12:53.000 --> 12:56.000]  If I visualize it, it happens at the same time.
[12:56.000 --> 13:01.000]  If I over-enode, I will see it on the three different versions.
[13:01.000 --> 13:07.000]  And then if I zoom a little bit, I guess we can see that...
[13:07.000 --> 13:09.000]  Wow.
[13:09.000 --> 13:11.000]  What can we see?
[13:11.000 --> 13:18.000]  We can see that PharmaSoft, for instance, is very connected most globally,
[13:18.000 --> 13:24.000]  but especially it has a very strong in-degree and not so big out-degree.
[13:24.000 --> 13:26.000]  Why is that?
[13:26.000 --> 13:31.000]  PharmaSoft is such a reference in France for open-source tools
[13:31.000 --> 13:34.000]  that it gets a lot of links from the whole community.
[13:34.000 --> 13:37.000]  And all websites of the Free Software community point to it,
[13:37.000 --> 13:40.000]  because it's like a resource.
[13:40.000 --> 13:46.000]  Whereas, of course, they cannot point to the whole rest of the community.
[13:46.000 --> 13:50.000]  On another note, I guess we could find...
[13:50.000 --> 13:53.000]  I think there was Linux...
[13:53.000 --> 13:56.000]  Linux-affair.org is the opposite.
[13:56.000 --> 14:03.000]  It's a media that pretty much talks about anything that happens on open-source in France.
[14:03.000 --> 14:07.000]  And, of course, they're the ones having the most outlinks.
[14:07.000 --> 14:10.000]  All right, so that's just a small example.
[14:10.000 --> 14:16.000]  Then I can show you maybe another notebook that will show other things.
[14:16.000 --> 14:21.000]  So this one is a notebook that was built out of data
[14:21.000 --> 14:24.000]  collected by Laura Miguel, which is a trainee at Media Lab right now.
[14:24.000 --> 14:29.000]  And she scraped the first-day website, the agendas,
[14:29.000 --> 14:35.000]  to try and get all speakers and rooms over the past 15 years.
[14:35.000 --> 14:40.000]  So here we will have to build the graph progressively.
[14:40.000 --> 14:46.000]  We just had a CSV that she scraped of the data of one speaker and one room.
[14:46.000 --> 14:49.000]  Disclaimer, the speakers have been anonymized.
[14:49.000 --> 14:53.000]  So you won't find a name that you know about,
[14:53.000 --> 14:57.000]  but they represent actual people.
[14:57.000 --> 15:01.000]  So let's take a look at, for instance, three examples of the data.
[15:01.000 --> 15:04.000]  So those are the three first lines.
[15:04.000 --> 15:08.000]  I mean, that's one line and two other lines that I picked specifically.
[15:08.000 --> 15:13.000]  This one is one speaker, and she talked about within this track.
[15:13.000 --> 15:17.000]  Here it was a stock that was shared between two speakers.
[15:17.000 --> 15:20.000]  So sometimes we get speakers separated by a pipe.
[15:20.000 --> 15:25.000]  And here is obviously someone that was still anonymized,
[15:25.000 --> 15:30.000]  but that should be in my seat right now.
[15:30.000 --> 15:35.000]  And we did many talks in the past, including in this room.
[15:35.000 --> 15:38.000]  So we will build the graph using NetworkX.
[15:38.000 --> 15:41.000]  So for those who know NetworkX, it's quite simple.
[15:41.000 --> 15:45.000]  You just create a new graph, and then for each row in our CSV,
[15:45.000 --> 15:49.000]  we will, if there's no speaker, we don't take it.
[15:49.000 --> 15:51.000]  Then we take the track and the year.
[15:51.000 --> 15:56.000]  We add a node for each track, and for each speaker inside the talk,
[15:56.000 --> 15:59.000]  we add a node for the speaker.
[15:59.000 --> 16:04.000]  And then we had an edge in between those two,
[16:04.000 --> 16:10.000]  and we increment it as a count if it's the second time that we meet him, for instance.
[16:10.000 --> 16:15.000]  And we also upgrade the year to get, for the edge,
[16:15.000 --> 16:18.000]  the last year that was used.
[16:18.000 --> 16:25.000]  So by doing that, I built a new graph that has 5,000 nodes and 6,000 links.
[16:25.000 --> 16:31.000]  Let's take a look at my alternate speaker here.
[16:31.000 --> 16:34.000]  It was supposed to be a speaker, and apparently,
[16:34.000 --> 16:41.000]  so it's linked to, in year 2018, to two talks in the graph room.
[16:41.000 --> 16:45.000]  Yeah, he spoke twice in the room, back then.
[16:45.000 --> 16:52.000]  In JavaScript in 2019, and in 2020 in the Open Research Tools and Technology Room.
[16:52.000 --> 16:55.000]  So let's take a look at this graph now.
[16:55.000 --> 16:56.000]  Oh, it was broken.
[16:56.000 --> 16:59.000]  Yes, there's a comma missing here.
[16:59.000 --> 17:00.000]  Here it is.
[17:00.000 --> 17:07.000]  Still, I tried to add this earlier, but I'm not expert enough with it, so I'll remove this.
[17:07.000 --> 17:09.000]  So here it is.
[17:09.000 --> 17:14.000]  So this time, it's a bipartite graph, since we got two kinds of nodes,
[17:14.000 --> 17:16.000]  the tracks and the speakers.
[17:16.000 --> 17:22.000]  So I decided that the node color will be attached to the part type.
[17:22.000 --> 17:31.000]  And if I take a look at it, we should see all big dots in blue are the rooms at first then,
[17:31.000 --> 17:34.000]  and all pink ones are actually speakers.
[17:34.000 --> 17:39.000]  And so we can see that there are a lot of lightning talks, of course, every year,
[17:39.000 --> 17:45.000]  but there are some rooms that have way more speakers than others,
[17:45.000 --> 17:48.000]  probably also because they exist for way longer.
[17:48.000 --> 17:53.000]  So maybe we can try and explore that, and that's the main idea.
[17:53.000 --> 17:58.000]  So sorry, I don't remember what this one is.
[17:58.000 --> 18:00.000]  Let's just run it briefly.
[18:00.000 --> 18:04.000]  I guess it's the same.
[18:04.000 --> 18:05.000]  Yeah, it's the same.
[18:05.000 --> 18:07.000]  Sorry, it's a copy-paste.
[18:07.000 --> 18:11.000]  So what we could do is try and apply other things.
[18:11.000 --> 18:13.000]  So let's do a grid again.
[18:13.000 --> 18:17.000]  But this time, we'll try and display for each node a gradient of color.
[18:17.000 --> 18:21.000]  That will indicate the intensity of the node at this moment.
[18:21.000 --> 18:29.000]  So to do that, we will, for instance, take a look at the year 2012 and the year 2022,
[18:29.000 --> 18:35.000]  and use the strength of the ALO, depending on how many talks were associated
[18:35.000 --> 18:37.000]  to this node for this specific year.
[18:37.000 --> 18:44.000]  So both graphs should show the intensity of the talks during those two years.
[18:44.000 --> 18:52.000]  So let's show it again.
[18:52.000 --> 18:58.000]  And here we can see that in 2012, the main rooms that were filled were actually more
[18:58.000 --> 19:01.000]  on desktops, Mozilla, Lightning Talks, and Embedded,
[19:01.000 --> 19:09.000]  whereas in 2022, there are way more rooms that are actually filled and spoken.
[19:09.000 --> 19:12.000]  Then what we could do is continue working on our graph
[19:12.000 --> 19:14.000]  and continue exploring while working with it.
[19:14.000 --> 19:17.000]  So at Medialab, we also have a tool called Pelot,
[19:17.000 --> 19:21.000]  which allows us to do a bunch of metrics and calculation on a,
[19:21.000 --> 19:23.000]  so it's already installed, it's going faster.
[19:23.000 --> 19:28.000]  And for instance, it can do a monopartite projection out of a bipartite graph.
[19:28.000 --> 19:33.000]  So I'm just running this, and then we can try and display it.
[19:33.000 --> 19:40.000]  And here, just in a few lines in Python, I can just see the alternate graph
[19:40.000 --> 19:43.000]  that is the monopartite version of the graph,
[19:43.000 --> 19:51.000]  and just see the links between the rooms depending on when they are co-spoken by speakers.
[19:51.000 --> 19:53.000]  Let's continue.
[19:53.000 --> 19:59.000]  And the problem is that if I look at this graph, I can see there are a bunch of nodes isolated.
[19:59.000 --> 20:04.000]  And so usually when I want to visualize a graph, those are a bit annoying
[20:04.000 --> 20:08.000]  because they take a lot of space in the visualization, and I don't want to see that.
[20:08.000 --> 20:13.000]  So let's just use Pelot's Crop to Largest Component function
[20:13.000 --> 20:19.000]  that will keep only the biggest component of the graph.
[20:19.000 --> 20:30.000]  So then I can re-displace this graph without all of those single nodes.
[20:30.000 --> 20:34.000]  And that's a rough idea of what could be done.
[20:34.000 --> 20:39.000]  Then we can work with the graph and just visualize on the fly.
[20:39.000 --> 20:46.000]  And I guess I'll conclude by just showing inside the GitHub page of the tool.
[20:46.000 --> 20:50.000]  There's all the visual variables that are available.
[20:50.000 --> 20:56.000]  So I showed you already node color, but you can also play on the saturation of the nodes.
[20:56.000 --> 21:04.000]  You can play on the size we saw, but you can play with the label size, of course, the label color.
[21:04.000 --> 21:10.000]  You can adjust the border ratio, how big it is.
[21:10.000 --> 21:23.000]  So basically all visual ways to better help you interpret your graph can be proposed.
[21:23.000 --> 21:29.000]  You can also add pictograms, use shapes for each node.
[21:29.000 --> 21:35.000]  You can use halos like I showed earlier.
[21:35.000 --> 21:40.000]  And play also a lot of those applied to edges.
[21:40.000 --> 21:46.000]  So you can play on the colors, the form of them, and so on and so on.
[21:46.000 --> 21:50.000]  And I guess that will be it for me.
[21:50.000 --> 21:53.000]  And I will take all of your questions.
[22:07.000 --> 22:12.000]  Sorry, I'm just scrolling back to things that are nicer.
[22:12.000 --> 22:13.000]  All of you.
[22:13.000 --> 22:14.000]  Yes.
[22:14.000 --> 22:23.000]  Can you preserve the layout between the different steps so you can execute the layout every time you go to a new cell and preserve it?
[22:23.000 --> 22:24.000]  That's a good question.
[22:24.000 --> 22:27.000]  I don't think it has been planned yet.
[22:27.000 --> 22:29.000]  Can you repeat the question?
[22:29.000 --> 22:30.000]  Yes, sorry.
[22:30.000 --> 22:40.000]  So the question was can we maintain the layout from one cell to the other and not having to re-click to apply the layout every time?
[22:40.000 --> 22:42.000]  I don't think so.
[22:42.000 --> 22:51.000]  And what I know is that the layout, the way Forza class works, has some chaos.
[22:51.000 --> 22:55.000]  But here it's always instantiated on the same seed.
[22:55.000 --> 23:00.000]  So whenever you run it, it will always generate the same exact layout.
[23:00.000 --> 23:02.000]  So that's something.
[23:02.000 --> 23:05.000]  But it won't reuse the one from the previous cells.
[23:05.000 --> 23:06.000]  No.
[23:06.000 --> 23:08.000]  That could be something that could be an idea.
[23:08.000 --> 23:09.000]  Yes?
[23:09.000 --> 23:12.000]  Do you have any numbers on the upper limits of this system?
[23:12.000 --> 23:15.000]  And the size of the graph that you're going to run here?
[23:15.000 --> 23:20.000]  So can you run the values of noting this one or the limits?
[23:20.000 --> 23:28.000]  So the question is about volume and amplitude and how big of a graph we can display with this.
[23:28.000 --> 23:32.000]  So I believe the limit is actually the one of your browser.
[23:32.000 --> 23:38.000]  So it will depend on your GPU and your CPU and your RAM.
[23:38.000 --> 23:51.000]  But I know that SigmaJS properly endows graphs with, I would say, 100,000 of nodes and links.
[23:51.000 --> 23:58.000]  I guess I know I already displayed one with a few million links and 100,000 of nodes.
[23:58.000 --> 24:03.000]  It takes a bit more time, of course.
[24:03.000 --> 24:08.000]  Do you support something like collapsing nodes and expanding them?
[24:08.000 --> 24:13.000]  For instance, in these kind of power graphs where the communities could collapse if you want to put height once,
[24:13.000 --> 24:17.000]  and they could be selectively expanded as well.
[24:17.000 --> 24:27.000]  So the question is, can we aggregate and split nodes that have, for instance, the same group?
[24:27.000 --> 24:33.000]  For me, it would be, I don't think it's built-in within Sigma for sure.
[24:33.000 --> 24:38.000]  Maybe in Pelot, the library I was showing, like the monopartite projection is pretty much this kind of ID.
[24:38.000 --> 24:44.000]  And I don't know, but it might be in Pelot.
[24:44.000 --> 24:48.000]  Yeah.
[24:48.000 --> 24:52.000]  You might try the GPU.
[24:52.000 --> 24:57.000]  Yeah, SigmaJS, sorry.
[24:57.000 --> 25:01.000]  So the question is, does this use the GPU to display the graph?
[25:01.000 --> 25:03.000]  Yes.
[25:03.000 --> 25:07.000]  SigmaJS is heavily relying on WebGL.
[25:07.000 --> 25:13.000]  The previous version of SigmaJS was proposing to choose between Canvas and WebGL.
[25:13.000 --> 25:18.000]  Right now, it's only WebGL, so it won't work with all browsers.
[25:18.000 --> 25:24.000]  But nowadays, most browsers know to work with the GPU.
[25:24.000 --> 25:28.000]  So, yes.
[25:28.000 --> 25:30.000]  Thank you so much.
[25:30.000 --> 25:49.000]  Thank you.