Average of confidence shuffling

[honghaoli42]

Original report by Honghao LI (Bitbucket: honghaoli42, ).

---

Currently the average is implemented as the average over exp(-Ixy), since exponential is convex, it underestimates the shuffled mutual information, leading to fewer cuts. I suggest using the average over Ixy instead, since the intended behavior of shuffled data is zero mutual information, rather than exp(-Ixy) == 1.

https://bitbucket.org/VCabeli/miic_r_package/src/b6a698e3c666c355d6df10c5732b84aab9acc17c/src/confidence_cut.cpp#lines-113