The answers for above and below are swapped in the generated dataset. This shouldn't affect the results of the paper (as it seems to be consistent across all partitions of the dataset), but I thought it would be good to fix that in the generator script.
The answers for above and below are swapped in the generated dataset. This shouldn't affect the results of the paper (as it seems to be consistent across all partitions of the dataset), but I thought it would be good to fix that in the generator script.