no code implementations • NeurIPS 2008 • Steven J. Phillips, Miroslav Dudík
For the generative case, we derive an entropy-based weighting that maximizes expected log likelihood under the worst-case true class proportions.