On 03/13/2014 11:21 PM, Alan Williams wrote:

Hi Tiago,

how do the blockmodel and modularity community functions behave with a graph with one giant component and a few relatively tiny disconnected components? Do I need to create a new graph of just the dominant component and then run the community functions on that, or will the functions lump the disconnected vertices into a block/community of the dominant component with no noticeable effect on the algorithm outcome?

Modularity maximization will always keep the components in different communities, even if they arise out of statistical fluctuations. Instead, blockmodel inference will look for statistical evidence, and components will be merged together into other blocks if there is not enough evidence to keep them separate. At this day and age, I would recommend against using modularity maximization for any purpose other than to show how such a bad idea it is. Best, Tiago -- Tiago de Paula Peixoto <tiago@skewed.de>