4. 朴素贝叶斯

一、朴素贝叶斯的基本方法

二、朴素贝叶斯参数估计

1.极大似然估计

2.贝叶斯估计

三、朴素贝叶斯算法

四、朴素贝叶斯代码实现

import numpy as np


# 加载数据的函数
def loaddata():
    X = np.array([[1, 'S'], [1, 'M'], [1, 'M'], [1, 'S'],
                  [1, 'S'], [2, 'S'], [2, 'M'], [2, 'M'],
                  [2, 'L'], [2, 'L'], [3, 'L'], [3, 'M'],
                  [3, 'M'], [3, 'L'], [3, 'L']])
    y = np.array([-1, -1, 1, 1, -1, -1, -1, 1, 1, 1, 1, 1, 1, 1, -1])
    return X, y


# 训练、计算各个概率值
def Train(trainset, train_labels):
    # 数据量
    m = trainset.shape[0]
    # 特征数
    n = trainset.shape[1]
    # 先验概率,key是类别值，value是类别的概率值
    prior_probability = {}
    # 条件概率，key的构造：类别，特征，特征值,value是
    conditional_probability = {}

    # 类别的可能取值
    labels = set(train_labels)
    # 计算先验概率,此时没有计算总数据量m
    for label in labels:
        prior_probability[label] = len(train_labels[train_labels == label])
    print('prior_probabilit =', prior_probability)

    # 计算条件概率
    for i in range(m):
        for j in range(n):
            # key的构造：类别，特征，特征值
            key = str(train_labels[i]) + ',' + str(j) + ',' + str(trainset[i][j])
            if key in conditional_probability:
                conditional_probability[key] += 1
            else:
                conditional_probability[key] = 1
    print('conditional_probability = ', conditional_probability)

    # 因字典在循环时不能改变，故定义新字典来保存值
    conditional_probability_final = {}
    for key in conditional_probability:
        # 取出当前的类别
        label = key.split(',')[0]
        conditional_probability_final[key] = conditional_probability[key] / prior_probability[int(label)]

    # 最终先验概率（除以总数据量m）
    for label in labels:
        prior_probability[label] = prior_probability[label] / m

    return prior_probability, conditional_probability_final, labels


# 定义预测函数
def predict(data):
    result = {}
    # 循环标签
    for label in train_labels_set:
        temp = 1.0
        for j in range(len(data)):
            key = str(label) + ',' + str(j) + ',' + str(data[j])
            # 条件概率连乘
            temp = temp * conditional_probability[key]
        # 在乘上先验概率
        result[label] = temp * prior_probability[label]
    print('result =', result)
    # 排序返回标签值
    return sorted(result.items(), key=lambda x: x[1], reverse=True)[0][0]


if __name__ == '__main__':
    X, y = loaddata()
    prior_probability, conditional_probability, train_labels_set = Train(X, y)
    print('conditional_probability = ', conditional_probability)

    # 预测
    y_hat = predict([2, 'S'])
    print('y_hat =', y_hat)

prior_probabilit = {1: 9, -1: 6}
conditional_probability =  {'-1,0,1': 3, '-1,1,S': 3, '-1,1,M': 2, '1,0,1': 2, '1,1,M': 4, '1,1,S': 1, '-1,0,2': 2, '1,0,2': 3, '1,1,L': 4, '1,0,3': 4, '-1,0,3': 1, '-1,1,L': 1}
conditional_probability =  {'-1,0,1': 0.5, '-1,1,S': 0.5, '-1,1,M': 0.3333333333333333, '1,0,1': 0.2222222222222222, '1,1,M': 0.4444444444444444, '1,1,S': 0.1111111111111111, '-1,0,2': 0.3333333333333333, '1,0,2': 0.3333333333333333, '1,1,L': 0.4444444444444444, '1,0,3': 0.4444444444444444, '-1,0,3': 0.16666666666666666, '-1,1,L': 0.16666666666666666}
result = {1: 0.02222222222222222, -1: 0.06666666666666667}
y_hat = -1

五、拉普拉斯修正代码实现

import numpy as np


# 加载数据的函数
def loaddata():
    X = np.array([[1, 'S'], [1, 'M'], [1, 'M'], [1, 'S'],
                  [1, 'S'], [2, 'S'], [2, 'M'], [2, 'M'],
                  [2, 'L'], [2, 'L'], [3, 'L'], [3, 'M'],
                  [3, 'M'], [3, 'L'], [3, 'L']])
    y = np.array([-1, -1, 1, 1, -1, -1, -1, 1, 1, 1, 1, 1, 1, 1, -1])
    return X, y


# 训练、计算各个概率值
def Train(trainset, train_labels):
    # 数据量
    m = trainset.shape[0]
    # 特征数
    n = trainset.shape[1]
    # 先验概率,key是类别值，value是类别的概率值
    prior_probability = {}
    # 条件概率，key的构造：类别，特征，特征值,value是
    conditional_probability = {}

    # 类别的可能取值
    labels = set(train_labels)
    # 计算先验概率,此时没有计算总数据量m
    for label in labels:
        prior_probability[label] = len(train_labels[train_labels == label]) + 1
    print('prior_probabilit =', prior_probability)

    # 计算条件概率
    for i in range(m):
        for j in range(n):
            # key的构造：类别，特征，特征值
            key = str(train_labels[i]) + ',' + str(j) + ',' + str(trainset[i][j])
            if key in conditional_probability:
                conditional_probability[key] += 1
            else:
                conditional_probability[key] = 1
    print('conditional_probability = ', conditional_probability)

    # 因字典在循环时不能改变，故定义新字典来保存值
    conditional_probability_final = {}
    for key in conditional_probability:
        # 取出当前的类别
        label = key.split(',')[0]
        key1 = key.split(',')[1]
        Ni = len(set(trainset[:, int(key1)]))
        print(Ni)
        conditional_probability_final[key] = (conditional_probability[key] + 1) / (prior_probability[int(label)] +Ni)

    # 最终先验概率（除以总数据量m）
    for label in labels:
        prior_probability[label] = prior_probability[label] / (m + len(labels))

    return prior_probability, conditional_probability_final, labels


# 定义预测函数
def predict(data):
    result = {}
    # 循环标签
    for label in train_labels_set:
        temp = 1.0
        for j in range(len(data)):
            key = str(label) + ',' + str(j) + ',' + str(data[j])
            # 条件概率连乘
            temp = temp * conditional_probability[key]
        # 在乘上先验概率
        result[label] = temp * prior_probability[label]
    print('result =', result)
    # 排序返回标签值
    return sorted(result.items(), key=lambda x: x[1], reverse=True)[0][0]


if __name__ == '__main__':
    X, y = loaddata()
    prior_probability, conditional_probability, train_labels_set = Train(X, y)
    print('conditional_probability = ', conditional_probability)

    # 预测
    y_hat = predict([2, 'S'])
    print('y_hat =', y_hat)

prior_probabilit = {1: 10, -1: 7}
conditional_probability =  {'-1,0,1': 3, '-1,1,S': 3, '-1,1,M': 2, '1,0,1': 2, '1,1,M': 4, '1,1,S': 1, '-1,0,2': 2, '1,0,2': 3, '1,1,L': 4, '1,0,3': 4, '-1,0,3': 1, '-1,1,L': 1}
conditional_probability =  {'-1,0,1': 0.4, '-1,1,S': 0.4, '-1,1,M': 0.3, '1,0,1': 0.23076923076923078, '1,1,M': 0.38461538461538464, '1,1,S': 0.15384615384615385, '-1,0,2': 0.3, '1,0,2': 0.3076923076923077, '1,1,L': 0.38461538461538464, '1,0,3': 0.38461538461538464, '-1,0,3': 0.2, '-1,1,L': 0.2}
result = {1: 0.027845457709711106, -1: 0.04941176470588235}
y_hat = -1

六、朴素贝叶斯处理连续数据

七、总结