posterior probability zero for all outcomes with exact methods

[sumanta]

test.net.gz

(5.96 KiB) Downloaded 348 times

Hi,


I am getting strange behaviour while trying to calculate the posterior probabilities
for the attached net file. for some nodes the posterior probability is zero for all
outcomes. Smile throws no error messages !

I'm using smile in a Ubuntu 16.04 (gcc 5.4 x64). I do a query for the node g24 in the
attached net file

test.net.gz

(5.96 KiB) Downloaded 348 times

.
I follow the steps in the Smile tutorial below:
-------------------------------------------------------------------------------------------
BN.SetDefaultBNAlgorithm(DSL_ALG_BN_LAURITZEN);
int success =BN.UpdateBeliefs();
if(success!=DSL_OKAY) printf("Inference Failed\n");
PrintAllPosteriors(BN);
-------------------------------------------------------------------------------------------


result:
...
...
P(g23=zero)=0.79312
P(g23=one)=0.097452
P(g23=two)=0.097452
P(g23=three)=0.0119761
P(g24=zero)=0
P(g24=one)=0
....
....



Strangely no error messages, and the inference returns DSL_OKAY.
Sampling algorithms seem to work.
The PrintAllPosteriors function is the same as in the tutorial.


Thanks in Advance for your help,


Sumanta

[shooltz[BayesFusion]]

I'm getting error -42 (DSL_OUT_OF_MEMORY) for the network you've attached with no evidence/no targets.