[SOLVED] Weighted Adjacency Matrix,Community Detection, GDF

[shafique]

Hi,
I have no experience of script writing and just switched from Pajek to Gephi 0.7 due to the need for detecting communities from a weighted adjacency matrix of 74 nodes. Somehow, my .NET file of Pajek did not open in Gephi. After failure with .dl type as well, I resorted to .gdf file.

In my matrix every element represents the frequency of co-occurance of corresponding nodes, and hence, it is the weight of the edge. When I open the .gdf file in Gephi, it gives only one error:

"Column type is not found for 'weight', string is used instead.
Edge property found: weight"

All the data loads neatly in Nodes and Edges tables of Gephi. However, the problem is that the graph shows every node connected to every other node which obviously is not the case. Moreover, the modularity algorithm lumps all the nodes in a single community while even a cursory view of my matrix and Pajek network suggests a strong likelihood that there are a few communities.

coclass_edges_ep_GDF.csv

Data File

(62.83KiB)Downloaded 456 times

The data file is attached in .csv format which I open in Notepad and save as .gdf type.

Can anybody help in pointing out what I may be doing wrong?

Moreover, what are other alternative softwares for community detection?

Any help will be highly appreciated.

Thanks in anticipation.

BR,
Shafique

[eduramiba]

Hi shafique,

Well I think you were missing the types for the attributes.

For example you need

Code: Select all

nodedef> name VARCHAR

instead of just

Code: Select all

nodedef> name

With these changes it loads ok.

Anyway, your file has a lot of edges, basically every node is connected to every node so the file does not seem correct for what you need.

[admin]

Hi,

The community detection method implemented in Gephi is the Louvain method: https://sites.google.com/site/findcommunities/
I don't think that it takes edge weight into account, sorry.

[WanderingAengus]

I'm trying to decide if Gephi is the tool I'm looking for so apologies for a n00b question and re-opening the thread.

I have the same situation: square, weighted adjacency matrix, but a little more interested in visualization than community detection. Is it possible to scale the edges to correspond to the edge weights? I probably wandered past this in the documentation, and didn't realize what I was looking at.

Once I know I can get there, I'll download/install/frantically play with Gephi to get something on the screen.

Thanks in advance.

[admin]

Edge thickness is linearly proportional to the edge weight.

[WanderingAengus]

Excellent! Now I just need to write a script to get the data from matrix to list format and I'm set. Thanks for the quick response! Cheers.

[admin]

You can use the DL file format to import a matrix:

http://gephi.org/users/supported-graph- ... dl-format/

[WanderingAengus]

Yep, I had picked that up from another post in the forum. Cheers.