Re: Random walks

Oskar Sandberg wrote:
> (Sorry for calling you Roger. :-) )

No worries. :-)

> I'm not sure it makes much sense to use one MH chain inside another, 
> though. We are already doing one "accept if r < min(1, ... )" step, so 
> an option is to multiply the above accept probability (deg(x)/deg(y)) 
> into that one when choosing whether or not accept the change.

My only concern with doing it that way would be efficiency: if 
unmetropolized walks tend to end at high-degree nodes, will we end up 
needing to send a large number of swap requests to get one accepted?

Cheers,
Michael