RegressionTeacher< BaseClassifierTeacher, RegressionClassifierOptions > Class Template Reference

Core class for the training of regression methods based on binary classifiers. More...

#include <Regression.h>

Inheritance diagram for RegressionTeacher< BaseClassifierTeacher, RegressionClassifierOptions >:

Collaboration diagram for RegressionTeacher< BaseClassifierTeacher, RegressionClassifierOptions >:

Public Member Functions
	RegressionTeacher (const GeneralOptions &general_options, const RegressionClassifierOptions &specific_options)
	Create a new teacher out of the general and the regression method specific options.
Weightfile	train (Dataset &training_data) const override
	Call the train function. More...

Protected Attributes
GeneralOptions	m_general_options
	GeneralOptions containing all shared options.

Private Attributes
RegressionClassifierOptions	m_specific_options
	The method specific options.
BaseClassifierTeacher	m_baseClassifierTeacher
	The teacher to use for training.

Detailed Description

template<class BaseClassifierTeacher, class RegressionClassifierOptions>
class Belle2::MVA::RegressionTeacher< BaseClassifierTeacher, RegressionClassifierOptions >

Core class for the training of regression methods based on binary classifiers.

The idea is as follows:

• If you want to solve a regression task, you need to make sure your target is between 0 and 1 and is more or less equally distributed (this is your task)
• The teacher will then train N binary classifiers, where N is given by the maximal bin number in the options.

• Each binary classifier is trained on the same data, but as a (binary) target is using

  s >= (1/number of bins) * (bin number)

  for every bin in [1, number of bins). This means, the first method is able to decide if a target value is very low or not, the second if it is medium low or not ... and the last if it is very high or not. The expert answer to a new data sample is the averaged output of all classifiers, which is very high, if all classifiers think it has a high value (then the target value is probably high) and very small, if non of them thinks the target value is high.

The class is very generic and can be used with every binary classifier.

Template Parameters

BaseClassifierTeacher	The teacher of the base binary classifier
RegressionClassifierOptions	The options of the classifier (probably inheriting from RegressionOptions)

Definition at line 139 of file Regression.h.

Member Function Documentation

train()

Belle2::MVA::Weightfile train ( Dataset &  training_data ) const

overridevirtual

Call the train function.

For each bin number in [1, number of bins) train a new classifier on the target

s >= (1/number of bins) * (bin number)

where s is the target variable of the dataset. By this, we end up with (number of bins) classifier, each trained on the same data but with different targets.

Implements Teacher.

Definition at line 245 of file Regression.h.

{
  Weightfile weightfile;
  weightfile.addOptions(m_general_options);
  weightfile.addOptions(m_specific_options);
 
  for (unsigned int binNumber = 1; binNumber < m_specific_options.getMaximalBinNumber(); binNumber++) {
    RegressionDataSet specificDataSet(m_general_options, &training_data,
                                      1.0 / m_specific_options.getMaximalBinNumber() * binNumber);
    Weightfile specificWeightFile = m_baseClassifierTeacher.train(specificDataSet);
 
    std::string file = weightfile.generateFileName();
    Weightfile::saveToXMLFile(specificWeightFile, file);
    weightfile.addFile("BaseClassifier_WeightFile_" + std::to_string(binNumber), file);
  }
  return weightfile;
}

---

The documentation for this class was generated from the following file:

• mva/methods/include/Regression.h