Probs modules are calculating negative Mutual Information

[Johnnyboycurtis]

See example:

frame
Out[8]: 
        U     V        MI
0   npreg   age  0.127532
1    skin   bmi  0.061203
2      bp   age  0.009789
3      bp   bmi  0.005360
4   npreg  skin  0.003926
5      bp  skin  0.003218
6     glu    bp  0.002185
7     glu   age  0.001108
8     glu   bmi  0.000553
9     glu  skin  0.000525
10  npreg   glu -0.000194
11   skin   age -0.000868
12  npreg    bp -0.001922
13    bmi   age -0.005869
14  npreg   bmi -0.012551

[Johnnyboycurtis]

A potential solution may be

from scipy.stats import chi2_contingency

def calc_MI(x, y, bins):
    c_xy = np.histogram2d(x, y, bins)[0]
    g, p, dof, expected = chi2_contingency(c_xy, lambda_="log-likelihood")
    mi = 0.5 * g / c_xy.sum()
    return mi

Taken from https://stackoverflow.com/questions/20491028/optimal-way-to-compute-pairwise-mutual-information-using-numpy

Another reference is:
https://en.wikipedia.org/wiki/G-test#Relation_to_Kullback–Leibler_divergence