朴素贝叶斯之简介

import numpy as np
from sklearn.naive_bayes import MultinomialNB
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score

# 1. 概率计算示例
print("=== 概率计算示例 ===")

# 数据：女神喜欢情况
data = [
    ['程序员', '喜欢'],
    ['产品经理', '喜欢'],
    ['程序员', '喜欢'],
    ['设计师', '喜欢'],
    ['程序员', '不喜欢'],
    ['产品经理', '不喜欢'],
    ['设计师', '不喜欢']
]

# 统计
total = len(data)
like_count = sum(1 for _, l in data if l == '喜欢')
not_like_count = total - like_count

print(f"总样本数: {total}")
print(f"喜欢的数量: {like_count}")
print(f"不喜欢的数量: {not_like_count}")
print(f"喜欢的概率: {like_count}/{total} = {like_count/total:.4f}")

# 条件概率：喜欢的情况下是程序员的概率
like_programmer = sum(1 for job, l in data if l == '喜欢' and job == '程序员')
print(f"\n喜欢且是程序员的数量: {like_programmer}")
print(f"喜欢的情况下是程序员的概率: {like_programmer}/{like_count} = {like_programmer/like_count:.4f}")

# 联合概率：喜欢且是程序员
print(f"联合概率（喜欢且是程序员）: {like_programmer}/{total} = {like_programmer/total:.4f}")

# 2. 朴素贝叶斯分类
print("\n=== 朴素贝叶斯分类 ===")

# 生成文本特征数据（词频矩阵）
from sklearn.feature_extraction.text import CountVectorizer

# 模拟商品评论
texts = [
    "这个商品很好 很喜欢",
    "质量不错 值得购买",
    "非常好 用起来很舒服",
    "差评 质量很差",
    "不好用 浪费钱",
    "太差了 不推荐"
]
labels = [1, 1, 1, 0, 0, 0]  # 1:好评, 0:差评

# 词频向量化
vectorizer = CountVectorizer()
X = vectorizer.fit_transform(texts)
print(f"特征矩阵形状: {X.shape}")
print(f"特征名称: {vectorizer.get_feature_names_out()}")

# 划分数据集
X_train, X_test, y_train, y_test = train_test_split(
    X, labels, test_size=0.33, random_state=42
)

# 朴素贝叶斯分类
nb = MultinomialNB(alpha=1.0)  # alpha是拉普拉斯平滑系数
nb.fit(X_train, y_train)

# 预测
y_pred = nb.predict(X_test)
acc = accuracy_score(y_test, y_pred)
print(f"准确率: {acc:.4f}")

# 3. 拉普拉斯平滑演示
print("\n=== 拉普拉斯平滑演示 ===")

# 模拟数据：某个词在好评中出现0次
word_in_positive = 0  # 好评中出现0次
word_in_negative = 5  # 差评中出现5次
total_positive = 10
total_negative = 10
alpha = 1  # 平滑系数
K = 2  # 类别数

# 不使用平滑
if word_in_positive == 0:
    print("不使用平滑: P(词|好评) = 0 (会导致整个概率为0)")

# 使用拉普拉斯平滑
p_word_positive = (word_in_positive + alpha) / (total_positive + alpha * K)
p_word_negative = (word_in_negative + alpha) / (total_negative + alpha * K)
print(f"使用平滑: P(词|好评) = {p_word_positive:.4f}")
print(f"使用平滑: P(词|差评) = {p_word_negative:.4f}")

# 4. 完整的情感分析示例
print("\n=== 完整情感分析示例 ===")

# 训练数据
train_texts = [
    "这个商品很好用 质量不错",
    "非常喜欢 推荐购买",
    "性价比高 值得入手",
    "质量好 物流快",
    "差评 质量太差了",
    "不好用 浪费钱",
    "太差了 不推荐",
    "失望 退货了"
]
train_labels = [1, 1, 1, 1, 0, 0, 0, 0]

# 测试数据
test_texts = [
    "质量很好 推荐",
    "太差了 退货"
]
test_labels = [1, 0]

# 词频向量化
vectorizer = CountVectorizer()
X_train = vectorizer.fit_transform(train_texts)
X_test = vectorizer.transform(test_texts)

# 朴素贝叶斯
nb = MultinomialNB(alpha=1.0)
nb.fit(X_train, train_labels)

# 预测
y_pred = nb.predict(X_test)
acc = accuracy_score(test_labels, y_pred)
print(f"测试集准确率: {acc:.4f}")

# 预测概率
y_proba = nb.predict_proba(X_test)
for text, pred, proba in zip(test_texts, y_pred, y_proba):
    label = '好评' if pred == 1 else '差评'
    print(f"'{text}' -> {label} (好评概率: {proba[1]:.4f})")

# 5. 不同alpha值的影响
print("\n=== 不同alpha值的影响 ===")
for alpha in [0.01, 0.1, 1.0, 10.0]:
    nb = MultinomialNB(alpha=alpha)
    nb.fit(X_train, train_labels)
    y_pred = nb.predict(X_test)
    acc = accuracy_score(test_labels, y_pred)
    print(f"alpha={alpha}: 准确率={acc:.4f}")