sqrt is applied twice on dists in median heuristic

[sanghack81]

dists = sqrt(dists)
sig = sqrt(0.5 * np.median(dists[dists > 0]))

Since dists is transformed once, there is no need apply sqrt on the median.
If X and Y in example.m is changed by multiplying 1000, I can see the outcome of tests is changed.

Should the second line be

sig = sqrt(0.5) * np.median(dists[dists > 0])

Once I change the second line, the test becomes immune to the scale of X or Y.