public BDeuScoringFunction(double ess, BayesianNetwork network, ADTree adTree, Constraints constraints)
{
this.network = network;
this.adTree = adTree;
this.constraints = constraints;
this.ess = ess;
for (int x = 0; x < network.Size(); x++)
{
Scratch s = new Scratch();
s.Variable = x;
s.Lri = Math.Log(network.GetCardinality(x));
s.invalidParents = new HashSet<ulong>();
scratchSpace.Add(s);
}
}
private void Lg(Varset parents, ref Scratch s)
{
int r = 1;
for (int pa = 0; pa < network.Size(); pa++)
{
if (parents.Get(pa))
{
r *= network.GetCardinality(pa);
}
}
s.Aij = ess / r;
s.Lgij = ScoreCalculator.GammaLn(s.Aij);
r *= network.GetCardinality(s.Variable);
s.Aijk = ess / r;
s.Lgijk = ScoreCalculator.GammaLn(s.Aijk);
}
private void Calculate(ContingencyTableNode ct, ulong currentBase, ulong index, Dictionary<ulong, int> paCounts, Varset variables, int previousVariable, ref Scratch s)
{
Varset variablesCp = new Varset(variables);
// if this is a leaf in the AD-tree
if (ct.IsLeaf())
{
// update the instantiation count of this set of parents
int count = paCounts.ContainsKey(index) ? paCounts[index] : 0;
count += ct.Value;
if (count > 0)
{
paCounts[index] = count;
// update the score for this variable, parent instantiation
double temp = ScoreCalculator.GammaLn(s.Aijk + ct.Value);
s.Score -= s.Lgijk;
s.Score += temp;
}
return;
}
// which actual variable are we looking at
int thisVariable = previousVariable + 1;
for (; thisVariable < network.Size(); thisVariable++)
{
if (variables.Get(thisVariable))
{
break;
}
}
// update the base and index if this is part of the parent set
ulong nextBase = currentBase;
if (thisVariable != s.Variable)
{
nextBase *= (ulong)network.GetCardinality(thisVariable);
}
// recurse
for (int k = 0; k < network.GetCardinality(thisVariable); k++)
{
ContingencyTableNode child = ct.GetChild(k);
if (child != null)
{
ulong newIndex = index;
if (thisVariable != s.Variable)
{
newIndex += currentBase * (ulong)k;
}
Calculate(child, nextBase, newIndex, paCounts, variables, thisVariable, ref s);
}
}
}