(a) UCB1 cumulative regret (mean of 200 runs) vs the bound
      T    regret     bound
    100      7.12    1290.9
   1000     51.36    1935.7
   5000    157.15    2386.3
  20000    315.99    2774.5

(b) tree search; correct leaf found at least once in N simulations / final most-visited path correct
  rho      N   independent sampling | binary: found  final path | graded: found  final path
 0.50     16                  0.223 |         0.140       0.000 |         0.205       0.000
 0.50     64                  0.635 |         0.335       0.000 |         0.470       0.000
 0.50    256                  0.982 |         0.625       0.270 |         1.000       0.850
 0.50   1024                  1.000 |         0.995       0.995 |         1.000       1.000
 0.40     16                  0.064 |         0.025       0.000 |         0.050       0.000
 0.40     64                  0.231 |         0.145       0.000 |         0.350       0.000
 0.40    256                  0.650 |         0.455       0.125 |         1.000       1.000
 0.40   1024                  0.985 |         0.980       0.980 |         1.000       1.000
 0.33     16                  0.022 |         0.020       0.000 |         0.025       0.000
 0.33     64                  0.084 |         0.065       0.000 |         0.235       0.000
 0.33    256                  0.296 |         0.315       0.110 |         1.000       1.000
 0.33   1024                  0.755 |         0.950       0.950 |         1.000       1.000
