utils_jensen_shannon.py 文件源码

def kullback_leibler_divergence(prob_dist1, prob_dist2, base=math.e):

    # Calculate the Kullback-Leibler divergence

    kl_divergence = 0

    # To avoid zero in the numerator or denominator
    pseudo_count = 0.000001

    for index in range(len(prob_dist1)):
        #print 'KL Divergence PD1[{0}]: {1} PD2[{0}]: {2}'.format(index, prob_dist1[index], prob_dist2[index])
        #print "newdiv == {0}".format(newdiv(float(prob_dist1[index]) + pseudo_count, float(prob_dist2[index]) + pseudo_count))
        #kl_divergence += prob_dist1[index] * math.log(newdiv(float(prob_dist1[index]) + pseudo_count, float(prob_dist2[index]) + pseudo_count), base)
        kl_divergence += prob_dist1[index] * math.log((float(prob_dist1[index]) + pseudo_count) / (float(prob_dist2[index]) + pseudo_count), base)

    return kl_divergence