Manupulating data in Potential Tables

[edj]

Hi guys

I am using unbbayes as API in my project. I would like to manupulate the data
in the Potential tables of a Bayesian Network in unbbayes. I having a problem
with how the data in Potential tables is organized. I can see that we can use
linear coordinates or multi-dimensional coordinates to access the data. I
thought I had an idea of how the multi-dimensional coordinates are organized
up until I tried to derive the linear coordinates from them. I suspected that
my problem could have emanated from not understanding how the states are
modelled. Are they represented by numbers from 0 up to n-1(where n is the
number of states a variable have). If possible can somebody help me with an
illustration, using example coordinates, how these coordinates can be
converted back and forth. Thank you.

[Rommel Novaes Carvalho]

Hi there,

I was searching on my gmail account to find the answer for your question. I
was sure I had answered this before! :-)

Here is a question I received via e-mail:

Rommel,

If i have to get the following ouput from he potential table (see values
below), what are the methods I should use

Potential table of Node C

state 0 of Node B and state 0 of Node C corresponds to value 0.5

state 0 of Node B and state 1 of Node C corresponds to value 0.3

state 1 of Node B and state 0 of Node C corresponds to value 0.5

state 1 of Node B and state 1 of Node C corresponds to value 0.7

Thanks.

--

Regards,

UnBBayes User

My answer was:

I am answering without checking the code (from the top of my head), so I
might get some methods wrong, but you will get the idea.

---

ProbabilisticNode node = ...;

ProbabilisticNode parentNode = ...;

PotentialTable table = ((PotentialTable)node.getProbabilisticFunction());

// To get B at state 1 and child node at state 1 (0.7) do

// number of the variables on the table = number of parents + 1 from the child
node

int numberOfVariables = 1 + 1;

int coord = new coord;

// coordinate of 0 is always the child node

coord = 1;

// for the parents you have to ask for the index

coord = 1;

float probability = table.getValueAt(coord);

---

As I said, I might have something wrong, but that is the overall idea.

I hope this answers your question!

Best,

Rommel

[Rommel Novaes Carvalho]

Summary:

ProbabilisticNode node = ...;
ProbabilisticNode parentNode = ...;
PotentialTable table = ((PotentialTable)node.getProbabilisticFunction());
// To get B at state 1 and child node at state 1 (0.7) do
// number of the variables on the table = number of parents + 1 from the child node
int numberOfVariables = 1 + 1;
int[] coord = new coord[numberOfVariables];
// coordinate of 0 is always the child node
coord[0] = 1;
// for the parents you have to ask for the index
coord[table.indexOfVariable(parentNode)] = 1;
float probability = table.getValueAt(coord);

I hope this looks better as a code.

Best,

Rommel