bitlch / bnt

Automatically exported from code.google.com/p/bnt
0 stars 0 forks source link

Missing Values Crash Learning In learn_params_dbn_em #28

Open GoogleCodeExporter opened 9 years ago

GoogleCodeExporter commented 9 years ago
What steps will reproduce the problem?
Running the following code will reproduce the problem:

intra = [0 1; 0 0];
inter = [1 0; 0 0];
num_nodes = 2;
num_states = [2 2];% num of states
dnodes = [1 2]; % indices of discrete nodes
onodes = 2;% indices of observed nodes
eclass1 = [1 2];
eclass2 = [3 2];

N = max([eclass1 eclass2]); 
CPT = cell(1,N);

bnet = mk_dbn(intra, inter, num_states, 'discrete', dnodes, ...
'observed', onodes, 'eclass1', eclass1, 'eclass2', eclass2);

bnet.CPD{1} = tabular_CPD(bnet,1);
bnet.CPD{2} = tabular_CPD(bnet,2);
bnet.CPD{3} = tabular_CPD(bnet,3);
engine = smoother_engine(jtree_2TBN_inf_engine(bnet));

ss = 2;%slice size(ss)
ncases = 10;%number of examples
T=10;
max_iter=2;%iterations for EM
cases = cell(1, ncases);
for i=1:ncases
  ev = sample_dbn(bnet, T);
  cases{i} = cell(ss,T);
  cases{i}(onodes,:) = ev(onodes, :);
  cases{i}{2,3} = [];
  cases{i}{2,4} = [];
end

[bnet2 LLTrace]= learn_params_dbn_em(engine, cases);

for i=1:N
  s=struct(bnet2.CPD{i});
  CPT{i}=s.CPT;
end
celldisp(CPT);

What is the expected output? What do you see instead?

I expect to see CPTs printed to the screen for the prior, transition, and 
emission probabilities. Instead, I get the following error:

Error using  .* 
Matrix dimensions must agree.

Error in mult_by_table (line 7)
bigT(:) = bigT(:) .* Ts(:); % must have bigT(:) on LHS to preserve
shape

Error in dpot/multiply_by_pot (line 11)
   Tbig.T = mult_by_table(Tbig.T, Tbig.domain, Tbig.sizes,
   Tsmall.T, Tsmall.domain, Tsmall.sizes);

Error in jtree_inf_engine/init_pot (line 17)
  clpot{c} = multiply_by_pot(clpot{c}, pots{i});

Error in jtree_2TBN_inf_engine/fwd (line 36)
[f.clpot, f.seppot] =  init_pot(engine.jtree_engine, clqs, pots,
engine.pot_type, engine.observed);

Error in smoother_engine/enter_evidence (line 14)
  [f{t}, ll(t)] = fwd(engine.tbn_engine, f{t-1}, ev(:,t), t);

Error in learn_params_dbn_em>EM_step (line 131)
  [engine, ll] = enter_evidence(engine, evidence);

Error in learn_params_dbn_em (line 82)
  [engine, loglik, logpost] = EM_step(engine, evidence,
  temperature);

Error in mem_no_aux (line 58)
bnet2 = learn_params_dbn_em( engine, cases);

What version of the product are you using? On what operating system?

I am using BNT Full Version 1.07 on 64 bit Windows 7.

Please provide any additional information below.

What has been provided is one of the sample HMM learning codes - it has been 
slightly augmented so that there are missing values in the samples for the 
observed nodes. Substituting a sample with the missing value symbol, [], causes 
element wise matrix multiplication errors in mult_by_table preventing the 
program from continuing.

Original issue reported on code.google.com by CaseyRHa...@gmail.com on 25 Jul 2012 at 1:02

GoogleCodeExporter commented 9 years ago
I have the same problem as you, I think the problem is related to the use of 
unobserved value symbol [].

Original comment by aqwzsx...@gmail.com on 21 Sep 2012 at 8:16

GoogleCodeExporter commented 9 years ago
See also, Issue 8, http://code.google.com/p/bnt/issues/detail?id=8
I patched my version and I don't get the error anymore.

Original comment by julian.k...@gmail.com on 3 Jun 2013 at 4:27