Another question is that, why do I have no option to set the degree in gephi?
What is the weight going to be used for?
[SOLVED] huge black blop from my gdf
-
equinox
- Posts:10
- Joined:23 Feb 2011 21:15 [phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1275: count(): Parameter must be an array or an object that implements Countable
Can someone please point me on what I need to do? I have a gdf file, but then when I import it to gephi, this is what I got
Another question is that, why do I have no option to set the degree in gephi?
What is the weight going to be used for?
Another question is that, why do I have no option to set the degree in gephi?
What is the weight going to be used for?
Last edited by equinox on 18 Apr 2011 05:44, edited 2 times in total.
Re: huge black blop from my gdf
lol.
Then, you select the "layout" panel and choose one option ("Force Atlas" is a favorite for many users, but it depends on your graph and your intentions). Then you click "run", and your big hairball should transform into something nicer.
Generally speaking, Gephi is not a software where you click "import" and get just one fixed visualization as a result. It has many functionalities (layout, but also filters, statistics, etc) which allow you to explore visually your data.
Best,
Clement
Then, you select the "layout" panel and choose one option ("Force Atlas" is a favorite for many users, but it depends on your graph and your intentions). Then you click "run", and your big hairball should transform into something nicer.
Generally speaking, Gephi is not a software where you click "import" and get just one fixed visualization as a result. It has many functionalities (layout, but also filters, statistics, etc) which allow you to explore visually your data.
Best,
Clement
Re: huge black blop from my gdf
Attached is a snapshot of your gdf example, imported in gephi and reworked the following way:
- run the modularity module in the stat tab
- partition according to the modularity class
- layout with openord (Force Atlas is too slow for your network, I did not realize it was so big)
- "expand" layout, x 4
- ranking for degree, rezise from 0 to 30.
The modularity score was low (0.2), when 0.4 is indicated as a minimum for being significant. So maybe that the two communities do not make real sense - you go back to the data and make your own opinion.
Also, your data has no attributes - which would have been useful for further exploration. Anyway!
Best,
Clement
- run the modularity module in the stat tab
- partition according to the modularity class
- layout with openord (Force Atlas is too slow for your network, I did not realize it was so big)
- "expand" layout, x 4
- ranking for degree, rezise from 0 to 30.
The modularity score was low (0.2), when 0.4 is indicated as a minimum for being significant. So maybe that the two communities do not make real sense - you go back to the data and make your own opinion.
Also, your data has no attributes - which would have been useful for further exploration. Anyway!
Best,
Clement
- Attachments
-
- with the ranking of nodes according to their degree
-
- Less messy gdf
-
equinox
- Posts:10
- Joined:23 Feb 2011 21:15 [phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1275: count(): Parameter must be an array or an object that implements Countable
Re: huge black blop from my gdf
Thank you very much for your help. I am pretty new to gephi and this really helps me a lot.seinecle wrote:Attached is a snapshot of your gdf example, imported in gephi and reworked the following way:
- run the modularity module in the stat tab
- partition according to the modularity class
- layout with openord (Force Atlas is too slow for your network, I did not realize it was so big)
- "expand" layout, x 4
- ranking for degree, rezise from 0 to 30.
The modularity score was low (0.2), when 0.4 is indicated as a minimum for being significant. So maybe that the two communities do not make real sense - you go back to the data and make your own opinion.
Also, your data has no attributes - which would have been useful for further exploration. Anyway!
Best,
Clement
How do you expand the layout x 4?
Can you explain what the modularity means?
When I try to partition, there is nothing in the drop down options...
I can't seem to find openord in the layout
I have updated my messy.gdf link above so that each nodes now have 5 attributes in it.
Re: huge black blop from my gdf
Hey,
So, to expand the layout, you actually choose a layout by this name in the layout table (called "expand"), and input 4 as a parameter in its parameter window.
For the partition, indeed sometimes the drop down menu is empty even when it should not. In this case, click on the "refresh" button (in the shape of two green arrows forming a circle) just on the left of the drop down menu, then check again the drop down menu - the attributes for partitioning should appear now.
"Modularity" is a function that you can run from inside the stats window, which launches the Louvain algorithm. This algorithm looks at the edges in the network, and detects the groups of nodes more densely interconnected by edges than they are connected with the rest of the network. Basically, it detects communities of nodes in the network. A detailed description of the steps followed by the algorithm is here:
http://iopscience.iop.org/1742-5468/200 ... P10008.pdf
In practice, this algo creates a new column of attribute for each node. The number in this column indicates to which community this node belongs, according to the communities detected by the algo ("0" = belongs to community 0, "1" = belongs to community 1, etc).
Finally, for the openord layout. I forgot that this layout is not included in the basic package of gephi. It is a plugin that you can download. For that, in the gephi menu, choose "Tools", then "Plugin", then select one of the tabs ("download" I think, but I am not so sure), and download and install the openord plugin. Once installed, it will appear in the list of available layouts in the layout tab.
I played a bit with your new gdf file with attributes. I could not find anything really striking, except for one anomaly: one of the nodes has a very high centrality (as measured by its pagerank, available in the stats tab). I resized the nodes so that the bigger the node, the higher the pagerank. But contrary to the rest of nodes with very high page rank, this big node has a very low reputation (as indicated by its green color, whereas other big nodes are red or blue, indicating middle to high reputations). How come that such a well connected node does not correlate with a high reputation?
(on the snapshot below, I selected this node so that its detail appear in the "edit" window on the top right.)
Best,
Clement
So, to expand the layout, you actually choose a layout by this name in the layout table (called "expand"), and input 4 as a parameter in its parameter window.
For the partition, indeed sometimes the drop down menu is empty even when it should not. In this case, click on the "refresh" button (in the shape of two green arrows forming a circle) just on the left of the drop down menu, then check again the drop down menu - the attributes for partitioning should appear now.
"Modularity" is a function that you can run from inside the stats window, which launches the Louvain algorithm. This algorithm looks at the edges in the network, and detects the groups of nodes more densely interconnected by edges than they are connected with the rest of the network. Basically, it detects communities of nodes in the network. A detailed description of the steps followed by the algorithm is here:
http://iopscience.iop.org/1742-5468/200 ... P10008.pdf
In practice, this algo creates a new column of attribute for each node. The number in this column indicates to which community this node belongs, according to the communities detected by the algo ("0" = belongs to community 0, "1" = belongs to community 1, etc).
Finally, for the openord layout. I forgot that this layout is not included in the basic package of gephi. It is a plugin that you can download. For that, in the gephi menu, choose "Tools", then "Plugin", then select one of the tabs ("download" I think, but I am not so sure), and download and install the openord plugin. Once installed, it will appear in the list of available layouts in the layout tab.
I played a bit with your new gdf file with attributes. I could not find anything really striking, except for one anomaly: one of the nodes has a very high centrality (as measured by its pagerank, available in the stats tab). I resized the nodes so that the bigger the node, the higher the pagerank. But contrary to the rest of nodes with very high page rank, this big node has a very low reputation (as indicated by its green color, whereas other big nodes are red or blue, indicating middle to high reputations). How come that such a well connected node does not correlate with a high reputation?
(on the snapshot below, I selected this node so that its detail appear in the "edit" window on the top right.)
Best,
Clement
- Attachments
-
- the middle sized green node: an oddity?
-
equinox
- Posts:10
- Joined:23 Feb 2011 21:15 [phpBB Debug] PHP Warning: in file [ROOT]/vendor/twig/twig/lib/Twig/Extension/Core.php on line 1275: count(): Parameter must be an array or an object that implements Countable
Re: huge black blop from my gdf
Clement,seinecle wrote:Hey,
So, to expand the layout, you actually choose a layout by this name in the layout table (called "expand"), and input 4 as a parameter in its parameter window.
For the partition, indeed sometimes the drop down menu is empty even when it should not. In this case, click on the "refresh" button (in the shape of two green arrows forming a circle) just on the left of the drop down menu, then check again the drop down menu - the attributes for partitioning should appear now.
"Modularity" is a function that you can run from inside the stats window, which launches the Louvain algorithm. This algorithm looks at the edges in the network, and detects the groups of nodes more densely interconnected by edges than they are connected with the rest of the network. Basically, it detects communities of nodes in the network. A detailed description of the steps followed by the algorithm is here:
http://iopscience.iop.org/1742-5468/200 ... P10008.pdf
In practice, this algo creates a new column of attribute for each node. The number in this column indicates to which community this node belongs, according to the communities detected by the algo ("0" = belongs to community 0, "1" = belongs to community 1, etc).
Finally, for the openord layout. I forgot that this layout is not included in the basic package of gephi. It is a plugin that you can download. For that, in the gephi menu, choose "Tools", then "Plugin", then select one of the tabs ("download" I think, but I am not so sure), and download and install the openord plugin. Once installed, it will appear in the list of available layouts in the layout tab.
I played a bit with your new gdf file with attributes. I could not find anything really striking, except for one anomaly: one of the nodes has a very high centrality (as measured by its pagerank, available in the stats tab). I resized the nodes so that the bigger the node, the higher the pagerank. But contrary to the rest of nodes with very high page rank, this big node has a very low reputation (as indicated by its green color, whereas other big nodes are red or blue, indicating middle to high reputations). How come that such a well connected node does not correlate with a high reputation?
(on the snapshot below, I selected this node so that its detail appear in the "edit" window on the top right.)
Best,
Clement
This actually made a lot of sense..this node called Community is a background process/admin of the whole website which manages other nodes, but doesn't take any credit for it, that explains the low reputation.
I was finally able to have the data visualized based on the instructions you provided above.
Thanks a bunch!
One thing that I would like to do with this data is to cluster the data based on age and location... mind helping me on how to proceed on with this?
Also what is the weight attribute of an edge used for this particular case? Is it just a measure of distance between nodes?
Re: huge black blop from my gdf
A few indications on the clustering by age and location:
Clustering works on the topology of the network (= the patterns traced by how edges and nodes connect). The attributes such as age, location etc. do not intervene in the process of community detection.
So, what is possible to do for age is:
- cluster your network with the Louvain algo.
- color-code the nodes according to the Louvain community they belong (you do that with the partitioning tab).
- resize the nodes according to age (with the ranking tab: the bigger the node, the oldest the node)
- visually inspect the resulting network: is one community (say, represented by red nodes) composed disproportionately of big nodes? Or is the distribution of size uniform across communities? (from what I could see the second option is the correct one). So I'd say that age does not correlate with the organizing of the topology of the network in communities. In other words, the age of the agents of your network does not seem to account for their clustering in communities.
For location, that's more tricky. I'll give you just the basic steps:
- outside gephi: use a geocoder service to convert the attribute "location" into latitude and longitude coordinates.
(http://developer.yahoo.com/maps/rest/V1/geocode.html)
(http://blog.jonudell.net/2007/08/10/exc ... dventures/)
(http://code.google.com/apis/maps/docume ... geocoding/)
[here you should be aware that the users you are interested may have provided fantasist answers, you'll have to filter that out etc.]
- I would also try to find a tool which would convert these coordinates into a country name! Could not find it with a quick search, though. Anyone?
- back to gephi: include these two new attributes, lat and long.
- use the geo layout plugin of gephi to map your nodes according to their spatial coordinates
(http://gephi.org/2010/map-geocoded-data-with-gephi/)
- partition / rank your nodes according to their age, reputation, etc... and get to see if it overlays in any manner with the geographical layout. If you succeeded in transforming your lat / long data into country names (again, I'd be very interested if anyone has a pointer to a web app which does that!), then you could:
* layout your network with openord, as before
* partition your nodes according to countries.
* viz the map: are nodes from particular countries close from each other?
Alternatively, actually even cooler:
* layout your network with openord, as before
* color rank your nodes according to their lat
* color rank your nodes according to their long (using the same colors as lat)
=> I did not try, but I suspect that if your nodes are structured according to a geographical logic (closer nodes in the network are also closer in their spatial location irl), your network will look like a heat map. If not (the network is a mash-up of colors), then it means that the location of the node does not contribute to the structure of the network. Which is the likely result I think, given the nature of your network.
Best,
Clement
Clustering works on the topology of the network (= the patterns traced by how edges and nodes connect). The attributes such as age, location etc. do not intervene in the process of community detection.
So, what is possible to do for age is:
- cluster your network with the Louvain algo.
- color-code the nodes according to the Louvain community they belong (you do that with the partitioning tab).
- resize the nodes according to age (with the ranking tab: the bigger the node, the oldest the node)
- visually inspect the resulting network: is one community (say, represented by red nodes) composed disproportionately of big nodes? Or is the distribution of size uniform across communities? (from what I could see the second option is the correct one). So I'd say that age does not correlate with the organizing of the topology of the network in communities. In other words, the age of the agents of your network does not seem to account for their clustering in communities.
For location, that's more tricky. I'll give you just the basic steps:
- outside gephi: use a geocoder service to convert the attribute "location" into latitude and longitude coordinates.
(http://developer.yahoo.com/maps/rest/V1/geocode.html)
(http://blog.jonudell.net/2007/08/10/exc ... dventures/)
(http://code.google.com/apis/maps/docume ... geocoding/)
[here you should be aware that the users you are interested may have provided fantasist answers, you'll have to filter that out etc.]
- I would also try to find a tool which would convert these coordinates into a country name! Could not find it with a quick search, though. Anyone?
- back to gephi: include these two new attributes, lat and long.
- use the geo layout plugin of gephi to map your nodes according to their spatial coordinates
(http://gephi.org/2010/map-geocoded-data-with-gephi/)
- partition / rank your nodes according to their age, reputation, etc... and get to see if it overlays in any manner with the geographical layout. If you succeeded in transforming your lat / long data into country names (again, I'd be very interested if anyone has a pointer to a web app which does that!), then you could:
* layout your network with openord, as before
* partition your nodes according to countries.
* viz the map: are nodes from particular countries close from each other?
Alternatively, actually even cooler:
* layout your network with openord, as before
* color rank your nodes according to their lat
* color rank your nodes according to their long (using the same colors as lat)
=> I did not try, but I suspect that if your nodes are structured according to a geographical logic (closer nodes in the network are also closer in their spatial location irl), your network will look like a heat map. If not (the network is a mash-up of colors), then it means that the location of the node does not contribute to the structure of the network. Which is the likely result I think, given the nature of your network.
Best,
Clement