@SnoopJ @rysiek @craigmaloney It's NOT a good approach. It pulls biases from the training data into the output.