Hirotaka NIITSUMA
2006-09-22
It is seem to EM algorithm and particle filer can be implemented by using
boost:: graph
And it gives more simple code.
May be other algorithm also can be implemented in same way.
Why boost:: graph is no used in OpenPNL.
Anonymous
2011-08-22
This is obviously in answer to a very old post but I've seen similar comments elsewhere.
It's not clear how the graph representation affects the EM algorithm which is independent of the implementation. It's also not clear how using BGL would simplify the code. One still needs to visit nodes and follow edges. That won't go away with BGL.
The internal graph representation used by PNL is a sparse adjacency matrix. I don't have the actual BGL code but the Quick Tour at the BGL site implies they use a dense matrix. Of course, this could be for illustration only. The internal representation only should affect access time. In what way would this be smaller with the BGL implementation? If a dense adjaceny matrix is used, it would require searching to collect the neighbors list. This would noticeably slow down any large network where any given nodes has only a small number of neighbors. There doesn't seem to be any real advantage to using BGL.