diff options
Diffstat (limited to 'sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map')
12 files changed, 516 insertions, 0 deletions
diff --git a/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/CVS/Entries b/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/CVS/Entries new file mode 100644 index 00000000..dc32f52b --- /dev/null +++ b/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/CVS/Entries @@ -0,0 +1,6 @@ +/disp_map_hhmm.m/1.1.1.1/Tue Sep 24 22:45:56 2002// +/learn_map.m/1.1.1.1/Sat Jan 11 18:48:46 2003// +/mk_map_hhmm.m/1.1.1.1/Tue Sep 24 10:49:52 2002// +/mk_rnd_map_hhmm.m/1.1.1.1/Tue Sep 24 22:13:48 2002// +/sample_from_map.m/1.1.1.1/Tue Sep 24 13:02:30 2002// +D/Old//// diff --git a/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/CVS/Repository b/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/CVS/Repository new file mode 100644 index 00000000..66b47bbc --- /dev/null +++ b/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/CVS/Repository @@ -0,0 +1 @@ +FullBNT/BNT/examples/dynamic/HHMM/Map diff --git a/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/CVS/Root b/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/CVS/Root new file mode 100644 index 00000000..f3bd14a6 --- /dev/null +++ b/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/CVS/Root @@ -0,0 +1 @@ +:ext:nsaunier@bnt.cvs.sourceforge.net:/cvsroot/bnt diff --git a/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/Old/CVS/Entries b/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/Old/CVS/Entries new file mode 100644 index 00000000..6079d451 --- /dev/null +++ b/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/Old/CVS/Entries @@ -0,0 +1,2 @@ +/mk_map_hhmm.m/1.1.1.1/Tue Sep 24 07:02:44 2002// +D diff --git a/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/Old/CVS/Repository b/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/Old/CVS/Repository new file mode 100644 index 00000000..354057a9 --- /dev/null +++ b/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/Old/CVS/Repository @@ -0,0 +1 @@ +FullBNT/BNT/examples/dynamic/HHMM/Map/Old diff --git a/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/Old/CVS/Root b/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/Old/CVS/Root new file mode 100644 index 00000000..f3bd14a6 --- /dev/null +++ b/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/Old/CVS/Root @@ -0,0 +1 @@ +:ext:nsaunier@bnt.cvs.sourceforge.net:/cvsroot/bnt diff --git a/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/Old/mk_map_hhmm.m b/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/Old/mk_map_hhmm.m new file mode 100644 index 00000000..7b646745 --- /dev/null +++ b/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/Old/mk_map_hhmm.m @@ -0,0 +1,156 @@ +function bnet = mk_map_hhmm(varargin) + +% p is the prob of a successful move (defines the reliability of motors) +p = 1; +num_obs_nodes = 1; + +for i=1:2:length(varargin) + switch varargin{i}, + case 'p', p = varargin{i+1}; + case 'numobs', num_obs_node = varargin{i+1}; + end +end + + +q = 1-p; + +% assign numbers to the nodes in topological order +U = 1; A = 2; C = 3; F = 4; O = 5; + +% create graph structure + +ss = 5; % slice size +intra = zeros(ss,ss); +intra(U,F)=1; +intra(A,[C F O])=1; +intra(C,[F O])=1; + +inter = zeros(ss,ss); +inter(U,[A C])=1; +inter(A,[A C])=1; +inter(F,[A C])=1; +inter(C,C)=1; + +% node sizes +ns = zeros(1,ss); +ns(U) = 2; % left/right +ns(A) = 2; +ns(C) = 3; +ns(F) = 2; +ns(O) = 5; % we will assign each state a unique symbol +l = 1; r = 2; % left/right +L = 1; R = 2; + +% Make the DBN +bnet = mk_dbn(intra, inter, ns, 'observed', O); +eclass = bnet.equiv_class; + + + +% Define CPDs for slice 1 +% We clamp all of them, i.e., do not try to learn them. + +% uniform probs over actions (the input could be chosen from a policy) +bnet.CPD{eclass(U,1)} = tabular_CPD(bnet, U, 'CPT', mk_stochastic(ones(ns(U),1)), ... + 'adjustable', 0); + +% uniform probs over starting abstract state +bnet.CPD{eclass(A,1)} = tabular_CPD(bnet, A, 'CPT', mk_stochastic(ones(ns(A),1)), ... + 'adjustable', 0); + +% Uniform probs over starting concrete state, modulo the fact +% that corridor 2 is only of length 2. +CPT = zeros(ns(A), ns(C)); % CPT(i,j) = P(C starts in j | A=i) +CPT(1, :) = [1/3 1/3 1/3]; +CPT(2, :) = [1/2 1/2 0]; +bnet.CPD{eclass(C,1)} = tabular_CPD(bnet, C, 'CPT', CPT, 'adjustable', 0); + +% Termination probs +CPT = zeros(ns(U), ns(A), ns(C), ns(F)); +CPT(r,1,1,:) = [1 0]; +CPT(r,1,2,:) = [1 0]; +CPT(r,1,3,:) = [q p]; +CPT(r,2,1,:) = [1 0]; +CPT(r,2,2,:) = [q p]; +CPT(l,1,1,:) = [q p]; +CPT(l,1,2,:) = [1 0]; +CPT(l,1,3,:) = [1 0]; +CPT(l,2,1,:) = [q p]; +CPT(l,2,2,:) = [1 0]; + +bnet.CPD{eclass(F,1)} = tabular_CPD(bnet, F, 'CPT', CPT); + + +% Assign each state a unique observation +CPT = zeros(ns(A), ns(C), ns(O)); +CPT(1,1,1)=1; +CPT(1,2,2)=1; +CPT(1,3,3)=1; +CPT(2,1,4)=1; +CPT(2,2,5)=1; +%CPT(2,3,:) undefined + +bnet.CPD{eclass(O,1)} = tabular_CPD(bnet, O, 'CPT', CPT); + + +% Define the CPDs for slice 2 + +% Abstract + +% Since the top level never resets, the starting distribution is irrelevant: +% A2 will be determined by sampling from transmat(A1,:). +% But the code requires we specify it anyway; we make it all 0s, a dummy value. +startprob = zeros(ns(U), ns(A)); + +transmat = zeros(ns(U), ns(A), ns(A)); +transmat(R,1,:) = [q p]; +transmat(R,2,:) = [0 1]; +transmat(L,1,:) = [1 0]; +transmat(L,2,:) = [p q]; + +% Qps are the parents we condition the parameters on, in this case just +% the past action. +bnet.CPD{eclass(A,2)} = hhmm2Q_CPD(bnet, A+ss, 'Fbelow', F, ... + 'startprob', startprob, 'transprob', transmat); + + + +% Concrete + +transmat = zeros(ns(C), ns(U), ns(A), ns(C)); +transmat(1,r,1,:) = [q p 0.0]; +transmat(2,r,1,:) = [0.0 q p]; +transmat(3,r,1,:) = [0.0 0.0 1.0]; +transmat(1,r,2,:) = [q p 0.0]; +transmat(2,r,2,:) = [0.0 1.0 0.0]; +% +transmat(1,l,1,:) = [1.0 0.0 0.0]; +transmat(2,l,1,:) = [p q 0.0]; +transmat(3,l,1,:) = [0.0 p q]; +transmat(1,l,2,:) = [1.0 0.0 0.0]; +transmat(2,l,2,:) = [p q 0.0]; + +% Add a new dimension for A(t-1), by copying old vals, +% so the matrix is the same size as startprob + + +transmat = reshape(transmat, [ns(C) ns(U) ns(A) 1 ns(C)]); +transmat = repmat(transmat, [1 1 1 ns(A) 1]); + +% startprob(C(t-1), U(t-1), A(t-1), A(t), C(t)) +startprob = zeros(ns(C), ns(U), ns(A), ns(A), ns(C)); +startprob(1,L,1,1,:) = [1.0 0.0 0.0]; +startprob(3,R,1,2,:) = [1.0 0.0 0.0]; +startprob(3,R,1,1,:) = [0.0 0.0 1.0]; +% +startprob(1,L,2,1,:) = [0.0 0.0 010]; +startprob(2,L,2,1,:) = [1.0 0.0 0.0]; +startprob(2,R,2,2,:) = [0.0 1.0 0.0]; + +% want transmat(U,A,C,At,Ct), ie. in topo order +transmat = permute(transmat, [2 3 1 4 5]); +startprob = permute(startprob, [2 3 1 4 5]); +bnet.CPD{eclass(C,2)} = hhmm2Q_CPD(bnet, C+ss, 'Fself', F, ... + 'startprob', startprob, 'transprob', transmat); + + diff --git a/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/disp_map_hhmm.m b/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/disp_map_hhmm.m new file mode 100644 index 00000000..0aadf2bb --- /dev/null +++ b/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/disp_map_hhmm.m @@ -0,0 +1,13 @@ +function disp_map_hhmm(bnet) + +eclass = bnet.equiv_class; +U = 1; A = 2; C = 3; F = 4; + +S = struct(bnet.CPD{eclass(A,2)}); +disp('abstract trans') +dispcpt(S.transprob) + +S = struct(bnet.CPD{eclass(C,2)}); +disp('concrete trans for go left') % UAC AC +dispcpt(squeeze(S.transprob(1,:,:,:,:))) + diff --git a/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/learn_map.m b/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/learn_map.m new file mode 100644 index 00000000..ac36586a --- /dev/null +++ b/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/learn_map.m @@ -0,0 +1,40 @@ +seed = 1; +rand('state', seed); +randn('state', seed); + +obs_model = 'unique'; % each cell has a unique label (essentially fully observable) +%obs_model = 'four'; % each cell generates 4 observations, NESW + +% Generate the true network, and a randomization of it +realnet = mk_map_hhmm('p', 0.9, 'obs_model', obs_model); +rndnet = mk_rnd_map_hhmm('obs_model', obs_model); +eclass = realnet.equiv_class; +U = 1; A = 2; C = 3; F = 4; onodes = 5; + +ss = realnet.nnodes_per_slice; +T = 100; +evidence = sample_dbn(realnet, 'length', T); +ev = cell(ss,T); +ev(onodes,:) = evidence(onodes,:); + +infeng = jtree_dbn_inf_engine(rndnet); + +if 0 +% suppose we do not observe the final finish node, but only know +% it is more likely to be on that off +ev2 = ev; +infeng = enter_evidence(infeng, ev2, 'soft_evidence_nodes', [F T], 'soft_evidence', {[0.3 0.7]'}); +end + + +learnednet = learn_params_dbn_em(infeng, {evidence}, 'max_iter', 5); + +disp('real model') +disp_map_hhmm(realnet) + +disp('learned model') +disp_map_hhmm(learnednet) + +disp('rnd model') +disp_map_hhmm(rndnet) + diff --git a/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/mk_map_hhmm.m b/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/mk_map_hhmm.m new file mode 100644 index 00000000..7b077ddb --- /dev/null +++ b/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/mk_map_hhmm.m @@ -0,0 +1,181 @@ +function bnet = mk_map_hhmm(varargin) + +% p is the prob of a successful move (defines the reliability of motors) +p = 1; +obs_model = 'unique'; + +for i=1:2:length(varargin) + switch varargin{i}, + case 'p', p = varargin{i+1}; + case 'obs_model', obs_model = varargin{i+1}; + end +end + + +q = 1-p; +unique_obs = strcmp(obs_model, 'unique'); + +% assign numbers to the nodes in topological order +U = 1; A = 2; C = 3; F = 4; +if unique_obs + onodes = 5; +else + N = 5; E = 6; S = 7; W = 8; % north, east, south, west + onodes = [N E S W]; +end + +% create graph structure + +ss = 4 + length(onodes); % slice size +intra = zeros(ss,ss); +intra(U,F)=1; +intra(A,[C F onodes])=1; +intra(C,[F onodes])=1; + +inter = zeros(ss,ss); +inter(U,[A C])=1; +inter(A,[A C])=1; +inter(F,[A C])=1; +inter(C,C)=1; + +% node sizes +ns = zeros(1,ss); +ns(U) = 2; % left/right +ns(A) = 2; +ns(C) = 3; +ns(F) = 2; +if unique_obs + ns(onodes) = 5; % we will assign each state a unique symbol +else + ns(onodes) = 2; +end +l = 1; r = 2; % left/right +L = 1; R = 2; + +% Make the DBN +bnet = mk_dbn(intra, inter, ns, 'observed', onodes); +eclass = bnet.equiv_class; + + + +% Define CPDs for slice 1 +% We clamp all the CPDs that are not tied, +% since we cannot learn them from a single sequence. + +% uniform probs over actions (the input could be chosen from a policy) +bnet.CPD{eclass(U,1)} = tabular_CPD(bnet, U, 'CPT', mk_stochastic(ones(ns(U),1)), ... + 'adjustable', 0); + +% uniform probs over starting abstract state +bnet.CPD{eclass(A,1)} = tabular_CPD(bnet, A, 'CPT', mk_stochastic(ones(ns(A),1)), ... + 'adjustable', 0); + +% Uniform probs over starting concrete state, modulo the fact +% that corridor 2 is only of length 2. +CPT = zeros(ns(A), ns(C)); % CPT(i,j) = P(C starts in j | A=i) +CPT(1, :) = [1/3 1/3 1/3]; +CPT(2, :) = [1/2 1/2 0]; +bnet.CPD{eclass(C,1)} = tabular_CPD(bnet, C, 'CPT', CPT, 'adjustable', 0); + +% Termination probs +CPT = zeros(ns(U), ns(A), ns(C), ns(F)); +CPT(r,1,1,:) = [1 0]; +CPT(r,1,2,:) = [1 0]; +CPT(r,1,3,:) = [q p]; +CPT(r,2,1,:) = [1 0]; +CPT(r,2,2,:) = [q p]; +CPT(l,1,1,:) = [q p]; +CPT(l,1,2,:) = [1 0]; +CPT(l,1,3,:) = [1 0]; +CPT(l,2,1,:) = [q p]; +CPT(l,2,2,:) = [1 0]; + +bnet.CPD{eclass(F,1)} = tabular_CPD(bnet, F, 'CPT', CPT); + + +% Observation model +if unique_obs + CPT = zeros(ns(A), ns(C), 5); + CPT(1,1,1)=1; % Theo state 4 + CPT(1,2,2)=1; % Theo state 5 + CPT(1,3,3)=1; % Theo state 6 + CPT(2,1,4)=1; % Theo state 9 + CPT(2,2,5)=1; % Theo state 10 + %CPT(2,3,:) undefined + O = onodes(1); + bnet.CPD{eclass(O,1)} = tabular_CPD(bnet, O, 'CPT', CPT); +else + % north/east/south/west can see wall (1) or opening (2) + CPT = zeros(ns(A), ns(C), 2); + CPT(:,:,1) = q; + CPT(:,:,2) = p; + bnet.CPD{eclass(W,1)} = tabular_CPD(bnet, W, 'CPT', CPT); + bnet.CPD{eclass(E,1)} = tabular_CPD(bnet, E, 'CPT', CPT); + CPT = zeros(ns(A), ns(C), 2); + CPT(:,:,1) = p; + CPT(:,:,2) = q; + bnet.CPD{eclass(S,1)} = tabular_CPD(bnet, S, 'CPT', CPT); + bnet.CPD{eclass(N,1)} = tabular_CPD(bnet, N, 'CPT', CPT); +end + +% Define the CPDs for slice 2 + +% Abstract + +% Since the top level never resets, the starting distribution is irrelevant: +% A2 will be determined by sampling from transmat(A1,:). +% But the code requires we specify it anyway; we make it all 0s, a dummy value. +startprob = zeros(ns(U), ns(A)); + +transmat = zeros(ns(U), ns(A), ns(A)); +transmat(R,1,:) = [q p]; +transmat(R,2,:) = [0 1]; +transmat(L,1,:) = [1 0]; +transmat(L,2,:) = [p q]; + +% Qps are the parents we condition the parameters on, in this case just +% the past action. +bnet.CPD{eclass(A,2)} = hhmm2Q_CPD(bnet, A+ss, 'Fbelow', F, ... + 'startprob', startprob, 'transprob', transmat); + + + +% Concrete + +transmat = zeros(ns(C), ns(U), ns(A), ns(C)); +transmat(1,r,1,:) = [q p 0.0]; +transmat(2,r,1,:) = [0.0 q p]; +transmat(3,r,1,:) = [0.0 0.0 1.0]; +transmat(1,r,2,:) = [q p 0.0]; +transmat(2,r,2,:) = [0.0 1.0 0.0]; +% +transmat(1,l,1,:) = [1.0 0.0 0.0]; +transmat(2,l,1,:) = [p q 0.0]; +transmat(3,l,1,:) = [0.0 p q]; +transmat(1,l,2,:) = [1.0 0.0 0.0]; +transmat(2,l,2,:) = [p q 0.0]; + +% Add a new dimension for A(t-1), by copying old vals, +% so the matrix is the same size as startprob + + +transmat = reshape(transmat, [ns(C) ns(U) ns(A) 1 ns(C)]); +transmat = repmat(transmat, [1 1 1 ns(A) 1]); + +% startprob(C(t-1), U(t-1), A(t-1), A(t), C(t)) +startprob = zeros(ns(C), ns(U), ns(A), ns(A), ns(C)); +startprob(1,L,1,1,:) = [1.0 0.0 0.0]; +startprob(3,R,1,2,:) = [1.0 0.0 0.0]; +startprob(3,R,1,1,:) = [0.0 0.0 1.0]; +% +startprob(1,L,2,1,:) = [0.0 0.0 010]; +startprob(2,L,2,1,:) = [1.0 0.0 0.0]; +startprob(2,R,2,2,:) = [0.0 1.0 0.0]; + +% want transmat(U,A,C,At,Ct), ie. in topo order +transmat = permute(transmat, [2 3 1 4 5]); +startprob = permute(startprob, [2 3 1 4 5]); +bnet.CPD{eclass(C,2)} = hhmm2Q_CPD(bnet, C+ss, 'Fself', F, ... + 'startprob', startprob, 'transprob', transmat); + + diff --git a/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/mk_rnd_map_hhmm.m b/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/mk_rnd_map_hhmm.m new file mode 100644 index 00000000..76b06fc7 --- /dev/null +++ b/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/mk_rnd_map_hhmm.m @@ -0,0 +1,73 @@ +function bnet = mk_rnd_map_hhmm(varargin) + +% We copy the deterministic structure of the real HHMM, +% but randomize the probabilities of the adjustable CPDs. +% The key trick is that 0s in the real HHMM remain 0 +% even when multiplied by a randon number. + +obs_model = 'unique'; + +for i=1:2:length(varargin) + switch varargin{i}, + case 'obs_model', obs_model = varargin{i+1}; + end +end + + +unique_obs = strcmp(obs_model, 'unique'); + +psuccess = 0.9; +% must be less than 1, so that pfail > 0 +% otherwise we copy too many 0s +bnet = mk_map_hhmm('p', psuccess, 'obs_model', obs_model); +ns = bnet.node_sizes; +ss = bnet.nnodes_per_slice; + +U = 1; A = 2; C = 3; F = 4; +%unique_obs = (bnet.nnodes_per_slice == 5); +if unique_obs + onodes = 5; +else + north = 5; east = 6; south = 7; west = 8; + onodes = [north east south west]; +end + +eclass = bnet.equiv_class; +S=struct(bnet.CPD{eclass(F,1)}); +CPT = mk_stochastic(rand(size(S.CPT)) .* S.CPT); +bnet.CPD{eclass(F,1)} = tabular_CPD(bnet, F, 'CPT', CPT); + + +% Observation model +if unique_obs + CPT = zeros(ns(A), ns(C), 5); + CPT(1,1,1)=1; % Theo state 4 + CPT(1,2,2)=1; % Theo state 5 + CPT(1,3,3)=1; % Theo state 6 + CPT(2,1,4)=1; % Theo state 9 + CPT(2,2,5)=1; % Theo state 10 + %CPT(2,3,:) undefined + O = onodes(1); + bnet.CPD{eclass(O,1)} = tabular_CPD(bnet, O, 'CPT', CPT); +else + for i=[north east south west] + CPT = mk_stochastic(rand(ns(A), ns(C), 2)); + bnet.CPD{eclass(i,1)} = tabular_CPD(bnet, i, 'CPT', CPT); + end +end + +% Define the CPDs for slice 2 + +startprob = zeros(ns(U), ns(A)); +S = struct(bnet.CPD{eclass(A,2)}); +transprob = mk_stochastic(rand(size(S.transprob)) .* S.transprob); +bnet.CPD{eclass(A,2)} = hhmm2Q_CPD(bnet, A+ss, 'Fbelow', F, ... + 'startprob', startprob, 'transprob', transprob); + +S = struct(bnet.CPD{eclass(C,2)}); +transprob = mk_stochastic(rand(size(S.transprob)) .* S.transprob); +startprob = mk_stochastic(rand(size(S.startprob)) .* S.startprob); +bnet.CPD{eclass(C,2)} = hhmm2Q_CPD(bnet, C+ss, 'Fself', F, ... + 'startprob', startprob, 'transprob', transprob); + + diff --git a/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/sample_from_map.m b/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/sample_from_map.m new file mode 100644 index 00000000..816b741e --- /dev/null +++ b/sourcecodes/bnt-master/BNT/examples/dynamic/HHMM/Map/sample_from_map.m @@ -0,0 +1,41 @@ +if 0 +% Generate some sample paths + +bnet = mk_map_hhmm('p', 1); +% assign numbers to the nodes in topological order +U = 1; A = 2; C = 3; F = 4; O = 5; + + +seed = 0; +rand('state', seed); +randn('state', seed); + +% control policy = sweep right then left +T = 10; +ss = 5; +ev = cell(ss, T); +ev(U,:) = num2cell([R*ones(1,5) L*ones(1,5)]); + +% fix initial conditions to be in left most state +ev{A,1} = 1; +ev{C,1} = 1; +evidence = sample_dbn(bnet, 'length', T, 'evidence', ev) + + +% Now do same but with noisy actuators + +bnet = mk_map_hhmm('p', 0.8); +evidence = sample_dbn(bnet, 'length', T, 'evidence', ev) + +end + +% Now do same but with 4 observations per slice + +bnet = mk_map_hhmm('p', 0.8, 'obs_model', 'four'); +ss = bnet.nnodes_per_slice; + +ev = cell(ss, T); +ev(U,:) = num2cell([R*ones(1,5) L*ones(1,5)]); +ev{A,1} = 1; +ev{C,1} = 1; +evidence = sample_dbn(bnet, 'length', T, 'evidence', ev) |