// compute the log likelihood.
double CLinearMapping::logLikelihood() const
{
double L=0.0;
for(unsigned int i=0; i<getNumData(); i++)
{
outActive.copyRowRow(0, b, 0);
outActive.gemvRowRow(0, W, *pX, i, 1.0, 1.0, "t");
L+=outActive.dist2Row(0, *py, i);
}
L=L/variance;
L+=(double)getNumData()*(ndlutil::LOGTWOPI + log(variance));
L*=-0.5;
//L+=priorLogProb();
return L;
}
void CLinearMapping::display(ostream& os) const
{
cout << "Linear Mapping:" << endl;
cout << "Optimiser: " << getDefaultOptimiserStr() << endl;
cout << "Data Set Size: " << getNumData() << endl;
cout << "Log likelihood: " << logLikelihood() << endl;
}
void hListBox::setScrollBarPosition(int start_item)
{
int numItems = getNumWidgets();
int numData = getNumData();
if(numData == numItems) return; // nothing to scroll
scrollBar->setPosition(start_item);
}
void CMlpMapping::display(ostream& os) const
{
cout << "Multi-Layer Perceptron Model:" << endl;
cout << "Optimiser: " << getDefaultOptimiserStr() << endl;
cout << "Data Set Size: " << getNumData() << endl;
cout << "Number hidden: " << hiddenDim << endl;
cout << "Log likelihood: " << logLikelihood() << endl;
}
// compute the log likelihood.
double CMlpMapping::logLikelihood() const
{
double L=0.0;
for(unsigned int i=0; i<getNumData(); i++)
{
hiddenActive.deepCopy(b1);
hiddenActive.gemvRowRow(0, W1, *pX, i, 1.0, 1.0, "t");
hiddenActive.tanh();
outActive.copyRowRow(0, b2, 0);
outActive.gemvRowRow(0, W2, hiddenActive, 0, 1.0, 1.0, "t");
L+=outActive.dist2Row(0, *py, i);
}
L=L/variance;
L+=(double)getNumData()*(ndlutil::LOGTWOPI + log(variance));
L*=-0.5;
//L+=priorLogProb();
return L;
}
void CLinearMapping::writeParamsToStream(ostream& out) const
{
writeToStream(out, "baseType", getBaseType());
writeToStream(out, "type", getType());
writeToStream(out, "numData", getNumData());
writeToStream(out, "outputDim", getOutputDim());
writeToStream(out, "inputDim", getInputDim());
writeToStream(out, "numParams", getOptNumParams());
CMatrix par(1, getOptNumParams());
getOptParams(par);
par.toStream(out);
}
// compute the gradients wrt parameters and latent variables.
double CLinearMapping::logLikelihoodGradient(CMatrix& g) const
{
double L=0.0;
g.zeros();
CMatrix gtemp(1, getOptNumParams());
for(unsigned int j=0; j<getOutputDim(); j++)
{
for(unsigned int i=0; i<getNumData(); i++)
{
double diff = outGradParams(gtemp, *pX, i, j) - py->getVal(i, j);
L+= diff*diff;
gtemp.scale(-diff/variance);
g.add(gtemp);
}
}
L=L/variance;
L+=(double)getNumData()*(ndlutil::LOGTWOPI + log(variance));
L*=-0.5;
//L+=priorLogProb();
return L;
}
