HMM实现中文分词

一时失言乱红尘 2021-10-19 17:34 320阅读 0赞

链接：[https://pan.baidu.com/s/1uBjLC61xm4tQ9raDa\_M1wQ][https_pan.baidu.com_s_1uBjLC61xm4tQ9raDa_M1wQ]  提取码：f7l1

推荐：[https://blog.csdn.net/longgb123/article/details/78154295][https_blog.csdn.net_longgb123_article_details_78154295]

import sys
    sys.path.append('保存文件的路径')  #设置路径
    # 下面三个文件在上面 
    
    from prob_emit import P as p_emit
    from prob_start import P as p_start
    from prob_trans import P as p_trans
    
    obs = '今天我来到北京清华大学'    #观察值
    states = 'BMES'
    
    V = [{}]
    path = {}
    
    prev_states = {
        'B': 'ES',  # t时刻：t-1时刻
        'M': 'BM',
        'E': 'BM',
        'S': 'SE'
    }
    
    # 初始化   第一个字符作为 'BMES' 的概率
    for y in states:
        V[0][y] = p_start[y] + p_emit[y][obs[0]]
        path[y] = y
    
    
    #概率： 输出概率(独立性概率：第k个字符状态为y的概率)  + 转换概率(y0 --> y ) + 上一个字符状态为y0的概率（当前状态与上一个字符的状态有关）
    
    for k in range(1, len(obs)):
        t0 = {}  # 储存概率
        path0 = {}  # 储存路径
        
        for y in states:
            em_p = p_emit[y][obs[k]]  # 输出概率   表示 第 k 个字符状态为 y 时 的输出概率
            
            
            (prob, s0) = max((em_p+p_trans[y0][y]+V[-1][y0], y0) for y0 in prev_states[y])
            
            # 动态规划
            #上面的prob  s0  就是在下面的每个循环中取tmp_prob tmp_s0 中取得最大值
    #         for y0 in prev_states[y]: # prev_state[y] 表示 t时刻 状态为 y 时的 t-1时刻可能出现的状态 
    #             tmp_prob = em_p + p_trans[y0][y] + V[-1][y0]
    #             tmp_s0 = y0
                
    
            t0[y] = prob
            path0[y] = path[s0] + y
    
        path = path0
        V.append(t0)
    path    
    (prob, s0) = max((V[-1][y], y) for y in 'ES')
    
    
    s2 = []
    for i, char in enumerate(obs):
        sign = path[s0][i]
        if(sign == 'B'):
            begin = i
        elif sign == 'E':
            s2.append(obs[begin: i+1])
        elif sign == 'S':
            s2.append(obs[i])
        else :
            pass
    s2

转载于:https://www.cnblogs.com/DDiamondd/p/11146388.html

[https_pan.baidu.com_s_1uBjLC61xm4tQ9raDa_M1wQ]: https://pan.baidu.com/s/1uBjLC61xm4tQ9raDa_M1wQ
[https_blog.csdn.net_longgb123_article_details_78154295]: https://blog.csdn.net/longgb123/article/details/78154295