Expert mixture Point operators

emo.cxx

Point ExpertMixture::evaluate_supervised(const Point & inP) const
{
  if (supervised_) return evaluate_supervised(inP);
  return evaluate_non_supervised(inP);
}

Point ExpertMixture::evaluate_non_supervised(const Point & inP) const
{
  const UnsignedInteger inputDimension = getInputDimension();
  if (inP.getDimension() != inputDimension) throw InvalidArgumentException(HERE) << "Error: expected a point of dimension=" << inputDimension << " and got a point of dimension=" << inP.getDimension();
  const UnsignedInteger outputDimension = getOutputDimension();
  const UnsignedInteger size = experts_.getSize();
  UnsignedInteger bestClass = 0;
  // Build the point z for the first class and grade it according to the classifier
  // In a non supervised case no need to stack data
  Point mixedPoint(inP);
  Scalar bestGrade = classifier_.grade(mixedPoint, bestClass);
  LOGDEBUG(OSS() << "Class index=" << 0 << ", grade=" << bestGrade);
  for (UnsignedInteger classIndex = 1; classIndex < size; ++classIndex)
  {
    // Build the point z for each other class and grade it according to the classifier
    const Scalar grade = classifier_.grade(mixedPoint, classIndex);
    LOGDEBUG(OSS() << "Class index=" << classIndex << ", grade=" << grade);
    // The best class will give the output value
    if (grade > bestGrade)
    {
      bestGrade = grade;
      bestClass = classIndex;
    }
    }
  const Point bestValue(experts_[bestClass](inP));
  LOGDEBUG(OSS() << "Best class index=" << bestClass << ", best grade=" << bestGrade << ", best value=" << bestValue);
  return bestValue;
}


Point ExpertMixture::evaluate_supervised(const Point & inP) const
{
  const UnsignedInteger inputDimension = getInputDimension();
  if (inP.getDimension() != inputDimension) throw InvalidArgumentException(HERE) << "Error: expected a point of dimension=" << inputDimension << " and got a point of dimension=" << inP.getDimension();
  const UnsignedInteger outputDimension = getOutputDimension();
  const UnsignedInteger size = experts_.getSize();
  UnsignedInteger bestClass = 0;
  // Build the point z for the first class and grade it according to the classifier
  Point mixedPoint(inP);
  // z=(x, f(x))
  Point bestValue(experts_[0](inP));
  mixedPoint.add(bestValue);
  Scalar bestGrade = classifier_.grade(mixedPoint, bestClass);
  LOGDEBUG(OSS() << "Class index=" << 0 << ", grade=" << bestGrade << ", value=" << bestValue);
  for (UnsignedInteger classIndex = 1; classIndex < size; ++classIndex)
  {
    // Build the point z for each other class and grade it according to the classifier
    const Point localValue(experts_[classIndex](inP));
    // z=(x, f(x))
    // mixedPoint is already defined so we just copy the localValue
    std::copy(localValue.begin(), localValue.end(), mixedPoint.begin() + inputDimension);
    const Scalar grade = classifier_.grade(mixedPoint, classIndex);
    LOGDEBUG(OSS() << "Class index=" << classIndex << ", grade=" << grade << ", value=" << localValue);
    // The best class will give the output value
    if (grade > bestGrade)
    {
      bestGrade = grade;
      bestClass = classIndex;
      bestValue = localValue;
    }
  }
  LOGDEBUG(OSS() << "Best class index=" << bestClass << ", best grade=" << bestGrade << ", best value=" << bestValue);
  return bestValue;
}