Naive Bayes - Probabilistic Classification

import numpy as np
import pandas as pd
from sklearn.naive_bayes import GaussianNB, MultinomialNB, BernoulliNB
from sklearn.feature_extraction.text import TfidfVectorizer, CountVectorizer
from sklearn.model_selection import cross_val_score, train_test_split
from sklearn.pipeline import Pipeline
from sklearn.datasets import load_breast_cancer
from sklearn.metrics import classification_report

# THREE NAIVE BAYES VARIANTS
nb_guide = {
    "GaussianNB":    "Continuous features following normal distribution (age, weight, measurements)",
    "MultinomialNB": "Count features (word frequencies in text, NLP bag-of-words)",
    "BernoulliNB":   "Binary features (word present/absent, yes/no flags)",
}
for variant, use_case in nb_guide.items():
    print(f"  {variant:16s}: {use_case}")

# GAUSSIAN NB ON MEDICAL DATA
cancer = load_breast_cancer()
X, y = cancer.data, cancer.target
gnb = GaussianNB()
cv = cross_val_score(gnb, X, y, cv=5, scoring="accuracy")
print(f"\nGaussianNB on breast cancer: CV={cv.mean():.4f} (trained in milliseconds!)")

# MULTINOMIAL NB FOR TEXT (SPAM DETECTION)
spam_emails = [
    "Special offer buy now click here amazing deal free money",
    "Act now limited time offer you are selected win prize",
    "Click here to claim your reward exclusive sale today",
    "Free vacation winner notification click immediately",
    "Hi John, can we schedule a meeting for Tuesday afternoon?",
    "Please review the attached quarterly report documents",
    "Team meeting agenda for next week project updates",
    "Your invoice for March services is attached for review",
    "Lunch at noon? Let me know if that works for the team",
    "Project deadline reminder please submit by Friday",
]
labels = [1, 1, 1, 1, 0, 0, 0, 0, 0, 0]  # 1=spam, 0=ham

X_train, X_test, y_train, y_test = train_test_split(spam_emails, labels, test_size=0.3, random_state=42)

spam_pipeline = Pipeline([
    ("tfidf",      TfidfVectorizer(stop_words="english", ngram_range=(1, 2))),
    ("classifier", MultinomialNB(alpha=1.0)),  # alpha=1: Laplace smoothing
])

spam_pipeline.fit(X_train, y_train)
y_pred = spam_pipeline.predict(X_test)
y_prob = spam_pipeline.predict_proba(X_test)[:, 1]

print("\nSpam Classifier:")
print(classification_report(y_test, y_pred, target_names=["Ham", "Spam"], zero_division=0))

# TEST NEW EMAILS
test_emails = [
    "Congratulations you won a prize click here",
    "Hi, let's discuss the project proposal tomorrow",
]
for email in test_emails:
    prob = spam_pipeline.predict_proba([email])[0, 1]
    verdict = "SPAM" if prob > 0.5 else "HAM"
    print(f"  '{email[:40]}...' -> {verdict} ({prob:.1%})")

# INCREMENTAL LEARNING -- unique to Naive Bayes
partial_gnb = GaussianNB()
X_arr, y_arr = cancer.data, cancer.target
# Update model with new data batches (streaming scenario)
for batch_start in range(0, len(X_arr), 50):
    partial_gnb.partial_fit(X_arr[batch_start:batch_start+50], y_arr[batch_start:batch_start+50], classes=np.unique(y_arr))
print(f"\nStreaming GaussianNB accuracy: {partial_gnb.score(X_arr, y_arr):.4f}")