Bernoulli graphs can have p>1 about fastrg HOT 7 CLOSED

Maybe we should switch the name of the edge distribution to "approximately-bernoulli" then too? And switch from an error to a warning? Does that sound good to you?

from fastrg.

from fastrg.

We can do things on a model by model basis. I think it makes sense to leave the error for model where the approximation is exact (i.e. above).

from fastrg.

But my probabilities weren't greater than one.... it only happens because of the -log(1-p) needed to inflate the Poisson probabilities... so that after thresholding, we get Bernoulli( p)

from fastrg.

So, the check should go before the -log(1-p)

from fastrg.

Ah I should move this code out of the SBM model and into the general factor model constructors.

from fastrg.

General solution in #29

from fastrg.