To check out this repository please hg clone the following URL, or open the URL using EasyMercurial or your preferred Mercurial client.
root / _FullBNT / BNT / general / Old / mk_gdl_graph.m @ 8:b5b38998ef3b
History | View | Annotate | Download (2.53 KB)
| 1 |
function gdl = mk_gdl_graph(G, domains, node_sizes, kernels, varargin) |
|---|---|
| 2 |
% MK_GDL_GRAPH Make a GDL (generalized distributed law) graph |
| 3 |
% gdl = mk_gdl_graph(G, domains, node_sizes, kernels, ...) |
| 4 |
% |
| 5 |
% A GDL graph is like a moralized, but untriangulated, Bayes net: |
| 6 |
% each "node" represents a domain with a corresponding kernel function. |
| 7 |
% For details, see "The Generalized Distributive Law", Aji and McEliece, |
| 8 |
% IEEE Trans. Info. Theory, 46(2): 325--343, 2000 |
| 9 |
% |
| 10 |
% G(i,j) = 1 if there is an (undirected) edge between domains i,j |
| 11 |
% |
| 12 |
% domains{i} is the domain of node i
|
| 13 |
% |
| 14 |
% node_sizes(i) is the number of values node i can take on, |
| 15 |
% or the length of node i if i is a continuous-valued vector. |
| 16 |
% node_sizes(i) = 1 if i is a utility node. |
| 17 |
% |
| 18 |
% kernels is the list of kernel functions |
| 19 |
% |
| 20 |
% The list below gives optional arguments [default value in brackets]. |
| 21 |
% |
| 22 |
% equiv_class - equiv_class(i)=j means factor node i gets its params from factors{j} [1:F]
|
| 23 |
% discrete - the list of nodes which are discrete random variables [1:N] |
| 24 |
% chance - the list of nodes which are random variables [1:N] |
| 25 |
% decision - the list of nodes which are decision nodes [ [] ] |
| 26 |
% utility - the list of nodes which are utility nodes [ [] ] |
| 27 |
|
| 28 |
|
| 29 |
ns = node_sizes; |
| 30 |
N = length(domains); |
| 31 |
vars = []; |
| 32 |
for i=1:N |
| 33 |
vars = myunion(vars, domains{i});
|
| 34 |
end |
| 35 |
Nvars = length(vars); |
| 36 |
|
| 37 |
gdl.equiv_class = 1:length(kernels); |
| 38 |
gdl.chance_nodes = 1:Nvars; |
| 39 |
gdl.utility_nodes = []; |
| 40 |
gdl.decision_nodes = []; |
| 41 |
gdl.dnodes = 1:Nvars; |
| 42 |
|
| 43 |
if nargin >= 5 |
| 44 |
args = varargin; |
| 45 |
nargs = length(args); |
| 46 |
for i=1:2:nargs |
| 47 |
switch args{i},
|
| 48 |
case 'equiv_class', bnet.equiv_class = args{i+1};
|
| 49 |
case 'chance', bnet.chance_nodes = args{i+1};
|
| 50 |
case 'utility', bnet.utility_nodes = args{i+1};
|
| 51 |
case 'decision', bnet.decision_nodes = args{i+1};
|
| 52 |
case 'discrete', bnet.dnodes = args{i+1};
|
| 53 |
otherwise, |
| 54 |
error(['invalid argument name ' args{i}]);
|
| 55 |
end |
| 56 |
end |
| 57 |
end |
| 58 |
|
| 59 |
|
| 60 |
gdl.G = G; |
| 61 |
gdl.vars = vars; |
| 62 |
gdl.doms = domains; |
| 63 |
gdl.node_sizes = node_sizes; |
| 64 |
gdl.cnodes = mysetdiff(vars, gdl.dnodes); |
| 65 |
gdl.kernels = kernels; |
| 66 |
gdl.type = 'gdl'; |
| 67 |
|
| 68 |
% Compute a bit vector representation of the set of domains |
| 69 |
% dom_bitv(i,j) = 1 iff variable j occurs in domain i |
| 70 |
gdl.dom_bitv = zeros(N, length(vars)); |
| 71 |
for i=1:N |
| 72 |
gdl.dom_bitv(i, domains{i}) = 1;
|
| 73 |
end |
| 74 |
|
| 75 |
% compute the interesection of the domains on either side of each edge (separating set) |
| 76 |
gdl.sepset = cell(N, N); |
| 77 |
gdl.nbrs = cell(1,N); |
| 78 |
for i=1:N |
| 79 |
nbrs = neighbors(G, i); |
| 80 |
gdl.nbrs{i} = nbrs;
|
| 81 |
for j = nbrs(:)' |
| 82 |
gdl.sepset{i,j} = myintersect(domains{i}, domains{j});
|
| 83 |
end |
| 84 |
end |
| 85 |
|
| 86 |
|