features.py

import math
from collections import Counter

# Extracting binary word features from text
def ext_feat(item):
    return set(item.lower().split())

# Computing mutual information between a binary feature and binary class label
def compute_mut_inf(feat_count, label_counts, joint_count, N):
    mi = 0.0
    for x in [0, 1]:
        for y in [0, 1]:
            p_xy = joint_count[(x, y)] / N
            if p_xy == 0:
                continue
            p_x = feat_count[x] / N
            p_y = label_counts[y] / N
            mi += p_xy * math.log2(p_xy / (p_x * p_y))
    return mi

# Calculating MI scores for all feature in the datasets
# labels:1 = spam, 0= ham 
def compute_feature_mut_inf(dataset, labels):
    N = len(dataset)
    mi_scores = {}
    all_feat = set().union(*dataset)

    for f in all_feat:
        feat_counts = Counter()
        label_counts = Counter()
        joint_counts = Counter()

        for feat, y in zip(dataset, labels):
            x = 1 if f in feat else 0
            feat_counts[x] += 1
            label_counts[y] += 1
            joint_counts[(x, y)] += 1

        mi_scores[f] = compute_mut_inf(
            feat_counts, label_counts, joint_counts, N
        )

    return mi_scores

# selecting features with MI score on a given threshold
def slt_feat(features, mi_scores, threshold=0.01):
    return {f for f in features if mi_scores.get(f, 0) >= threshold}