You mention that AdaBoostClassifier works only with model accepting this feature. I don’t know how the sample weight is working. But, for models that doesn’t have such a feature, could you duplicate the sample to make them more significative in the data set to adapt this algorithm to all models?
Duplicating a sample will force the sample weights to be integral values. The reweighted scheme in the original algorithm requires floating-point values because it is a real function. Therefore, we cannot use duplication.
1 Like