We need to ensure that the positive_class is consistent for binary classification between TabRepo and OpenML/AutoMLBenchmark. Maybe the positive class is noted in the OpenML metadata, but I'm not sure.