I see. To initialize the NestedBlockState I'm doing this:

bs = [np.zeros(1)] * 10

state = NestedBlockState(g=g, bs=bs, sampling=True, deg_corr=True)

Is there a better way to initialize it such that this process is faster? Running "minimize_nested_blockmodel_dl" once is seemingly out of the question because of memory usage.

Also, I notice that only one of my CPU cores is being utilized. Is there no multicore low-memory way of doing this?

Thanks!

On Wed, Feb 12, 2020 at 3:23 AM Tiago de Paula Peixoto <tiago@skewed.de> wrote:

Hi,

Am 11.02.20 um 21:30 schrieb Deklan Webster:
> Thanks, I got it setup. The memory usage is now about 2~GB!
>
> I was testing with a smaller graph and I noticed the sweeps are much
> slower with multiflip. Is this a fundamental tradeoff between speed and
> memory usage, or is there some setting to make it faster?

It's an illusion that multiflip is slower, it is in fact much master. It
may take slightly longer to return per sweep, but it moves much further
in state space. Overall, the sampling equilibrates much faster.

Best,
Tiago

--
Tiago de Paula Peixoto <tiago@skewed.de>
_______________________________________________
graph-tool mailing list
graph-tool@skewed.de
https://lists.skewed.de/mailman/listinfo/graph-tool