pairwiseprediction

Latest Release | Current Code | API Documentation

Handle a regression problem inducing the model on pairs A,B of instances.

Last column of X is the continuous target. y is ignored.

Four modes are hipotetically possible, although only mode 'b' below is provided by now to ease compatibility with sklearn:

• a classification algorithm is trained to tell when instance A has higher target than instance B
  • 1. the prediction is based on interpolation which is the type of result expected from a regression
  • 2. the interpolated value can be converted to a hard prediction through binarization
• a regression algorithm is trained to predict the difference between target values of A,B
  • 3. the prediction is the value provided directly by the regressor
  • 4. the regression value can be converted to a hard prediction through binarization