// virtual
void GGaussianProcess::predictDistribution(const GVec& in, GPrediction* out)
{
if(!m_pBuf)
m_pBuf = new GMatrix(2, m_pStoredFeatures->rows());
else if(m_pBuf->rows() < 2)
m_pBuf->newRow();
// Compute k*
GVec& k = m_pBuf->row(0);
for(size_t i = 0; i < m_pStoredFeatures->rows(); i++)
k[i] = m_weightsPriorVar * m_pKernel->apply(m_pStoredFeatures->row(i), in);
// Compute the prediction
GVec pred;
m_pAlpha->multiply(m_pBuf->row(0), pred, true);
// Compute the variance
GVec& v = m_pBuf->row(1);
m_pLInv->multiply(k, v);
double variance = m_pKernel->apply(k, k) - v.squaredMagnitude();
// Store the results
for(size_t i = 0; i < m_pAlpha->cols(); i++)
{
GNormalDistribution* pNorm = out->makeNormal();
pNorm->setMeanAndVariance(pred[i], variance);
}
}
// virtual
void GLinearDistribution::predictDistribution(const GVec& in, GPrediction* out)
{
m_pAInv->multiply(in, m_buf);
double v = in.dotProduct(m_buf);
for(size_t i = 0; i < m_pWBar->rows(); i++)
{
GNormalDistribution* pNorm = (*out).makeNormal();
double m = m_pWBar->row(i).dotProduct(in);
pNorm->setMeanAndVariance(m, v);
out++;
}
}
// virtual
void GNaiveInstance::predictDistribution(const GVec& pIn, GPrediction* pOut)
{
GVec::setAll(m_pWeightSums, 0.0, m_pRelLabels->size());
GVec::setAll(m_pValueSums, 0.0, m_pRelLabels->size());
for(size_t i = 0; i < m_pRelFeatures->size(); i++)
evalInput(i, pIn[i]);
for(size_t i = 0; i < m_pRelLabels->size(); i++)
{
GNormalDistribution* pNorm = pOut[i].makeNormal();
pNorm->setMeanAndVariance(m_pValueSums[i] / m_pWeightSums[i], 1.0 / m_pWeightSums[i]);
}
}
// virtual
void GLinearDistribution::predictDistributionInner(const double* pIn, GPrediction* pOut)
{
size_t dims = m_pAInv->rows();
m_pAInv->multiply(pIn, m_pBuf);
double v = GVec::dotProduct(pIn, m_pBuf, dims);
for(size_t i = 0; i < m_pWBar->rows(); i++)
{
GNormalDistribution* pNorm = (*pOut).makeNormal();
double m = GVec::dotProduct(m_pWBar->row(i), pIn, dims);
pNorm->setMeanAndVariance(m, v);
pOut++;
}
}
// virtual
void GInstanceTable::predictDistribution(const double* pIn, GPrediction* pOut)
{
size_t pos = 0;
for(int i = 0; i < m_dims; i++)
{
size_t n = (size_t)floor(pIn[i] + 0.5);
if(n >= m_pDims[i])
ThrowError("dim=", gformat(i), ", index=", gformat(pIn[i]), ", out of range. Expected >= 0 and < ", gformat(m_pDims[i]));
pos += n * m_pScales[i];
}
int values = m_pRelation->valueCount(m_dims);
if(values == 0)
{
GNormalDistribution* pNorm = pOut->makeNormal();
pNorm->setMeanAndVariance(m_pTable[pos], 1.0);
}
else
{
GCategoricalDistribution* pCat = pOut->makeCategorical();
pCat->setSpike(values, (int)m_pTable[pos], 1);
}
}
