epsilon = 1/log(3); self plan = [[3/8,1/8],[1/8,3/8]]
self: transport=0.250000000; KL=0.130812036; full objective=0.369070246
collapsed mean: raw=0.250000000; debiased=0.065464877; debiased self=0
epsilon | target | pure transport | full KL objective | negative entropy objective
0.100000000 | same_two_piles | 0.000045398 | 0.069310178 | -0.069319258
0.100000000 | one_pile_at_mean | 0.250000000 | 0.250000000 | 0.180685282
0.300000000 | same_two_piles | 0.034445196 | 0.197428429 | -0.218459879
0.300000000 | one_pile_at_mean | 0.250000000 | 0.250000000 | 0.042055846
0.910239227 | same_two_piles | 0.250000000 | 0.369070246 | -0.892789261
0.910239227 | one_pile_at_mean | 0.250000000 | 0.250000000 | -0.380929754
1.500000000 | same_two_piles | 0.339243631 | 0.418165641 | -1.661275901
1.500000000 | one_pile_at_mean | 0.250000000 | 0.250000000 | -0.789720771
3.000000000 | same_two_piles | 0.417429794 | 0.458524818 | -3.700358266
3.000000000 | one_pile_at_mean | 0.250000000 | 0.250000000 | -1.829441542
PASS: 462 checks; 144 independent 2x2 comparisons; 6 rejected inputs
