inference.py

import math
from features import ext_feat, slt_feat

#  Computes posterior probability of spam membership given Bloom filter match using Bayes' rule.
def post_mbshp_prob(bf, item, mi_scores, prior=0.1, threshold=0.01):
    selected = slt_feat(ext_feat(item), mi_scores, threshold)

    if not selected or not all(bf.contains(f) for f in selected):
        return 0.0

    p_fp = bf.theoretical_fpr()
    return prior / (prior + p_fp * (1 - prior))

# Computes binary conditional entropy of the posterior probability.
def cond_ent(posterior):
    if posterior <= 0.0 or posterior >= 1.0:
        return 0.0
    return -(posterior * math.log2(posterior)
             + (1 - posterior) * math.log2(1 - posterior))