no double declaration of edges

[ghost]

In kalman_example, we write

>> xin.setDest(e);
>> e.setConnections(xin, a, b);

There is something unsatisfying about this in the sense that the connection xin-to-e is declared twice. Actually, all the connections are declared twice in kalman_example.

Is it possible to declare connections just once by code like this (or alike):

>> link1 = ffg.Link( xin, e ); % declares link from xin to e
>> link2 = ffg.Link( e, b ): % equ node to add node

This has the advantage that the direction of the edge is clear ( edge = ffg.Link(FromNode, ToNode) ), and that the variables in the network (ie, the links) can be assigned to a MATLAB variable (link1, link2, etc).

Note sure how this would work out at the back-end but for the user the network consists of node objects and link objects. FIrst we declare the node objects and then the link objects, eg,

>> xin = ffg.EvidenceNode( 'property', 'value');
>> e = ffg.EqualityNode(); 
>> link1 = ffg.Link( xin, e ); % connects xin to e

[psycharo-zz]

I got your point, what I don't like is that the Link is not useful by itself, so it might be better to create e.g. ffg.Network, with a method addEdge(a, b) or smth. This approach should also be useful when implementing loopy mode, the network will be in charge of message scheduling

[ghost]

OK, this addEdge(a,b) idea works fine. Problem solved.

One question I have. What does the network class do other than group a few functions. Do you need a class for that? Why not have a toolbox with folder structure +ffg/core and put the functions makeStep, addEdge, setSchedule in the core folder. This is core functionality of the toolbox. Yuou can call these fucntions by `ffg.addEdge

[psycharo-zz]

it verifies (for now very basically) whether the schedule is correct or not. also the schedule itself should be stored somewhere.

[ghost]

ok, sounds good, thx