Am 01.03.21 um 17:19 schrieb ulgenklc:
Ahh, yes I noticed that when I scanned through the previous posts, sorry for replication. I have three more quick questions:
1) If I fitted a different SBM to splitted the graph, wouldn't the communities in individual layers be temporally discrete? That's not quite what I want because I want to track an evolution, but I'm open to process that result in a temporally overlapping fashion(some sort of set matching algorithms between layers.)
I'm not sure exactly what you mean, but if you want to "align" the different partitions you can take a look at https://graph-tool.skewed.de/static/doc/inference.html#graph_tool.inference.... There is an example of this alignment being done here: https://graph-tool.skewed.de/static/doc/demos/inference/inference.html#sampl...
2) When I pass overlap =True to LayeredBlockState, I have NxT many nodes as a result like you said, that's all fine. However, now I can't get the node membership using .get_blocks().
levels = states[0].get_levels()
levels[0].get_blocks() is a list of length M(total number of edges in the network) which should be a list of length NxT, isn't it? Again, it feels like I'm missing something very trivial here...
The overlapping block model consists of a labeling of the half-edges of the graph. Since each half-edge can belong to only one layer, you can then tell how the membership of the respective node has changed in the time slice. Please take a look at the documentation on how to extract this information, e.g. via https://graph-tool.skewed.de/static/doc/inference.html#graph_tool.inference....
3) Also, is there a method for levels[0] that I can call to get the total number of communities up front? I can see the number of blocks when I do states[0].print_summary() but I need the integer value of this number for preprocessing..
It's worthwhile to peruse the documentation, which contains all the answers to questions such as this: https://graph-tool.skewed.de/static/doc/inference.html#graph_tool.inference.... -- Tiago de Paula Peixoto <tiago@skewed.de>