基于统计的分词(维特比算法、马尔可夫算法)

# 维特比算法
def vtb(n, o, s, ps, pe, pt):
    ret = {}
    path = {}
    if n == 0:
        for x in s:
            ret[x] = ps[x] * pe[x][o[n]]
            path[x] = [x]
    else:
        lret, lp = vtb(n - 1, o, s, ps, pe, pt)  # n-1天的,结果有3^n-1个
        for x in s:
            ret[x], mlx = max((lret[lx] * pt[lx][x] * pe[x][o[n]], lx) for lx in s)
            path[x] = lp[mlx] + [x]
    return ret, path  # 返回最大值和路径


# 马尔科夫算法
def hmm(n, o, s, ps, pe, pt):
    ret = {}
    if n == 0:
        for x in s:
            ret[x] = ps[x] * pe[x][o[n]]
    else:
        lret = hmm(n - 1, o, s, ps, pe, pt)  # n-1天的,结果有3^n-1个
        for k, v in lret.items():
            for x in s:
                ret[k + "-" + x] = v * pt[k.split("-")[-1]][x] * pe[x][o[n]]
    return ret


if __name__ == '__main__':
    p_start = {'good': 0.2, 'normal': 0.6, 'bad': 0.2}  # 初始概率矩阵
    p_emit = {
        'good': {'working': 0.05, 'travel': 0.35, 'shopping': 0.35, 'running': 0.25},
        'normal': {'working': 0.25, 'travel': 0.25, 'shopping': 0.25, 'running': 0.25},
        'bad': {'working': 0.6, 'travel': 0.2, 'shopping': 0.05, 'running': 0.15}
    }  # 发射概率矩阵
    p_trans = {
        'good': {'good': 0.5, 'normal': 0.375, 'bad': 0.125},
        'normal': {'good': 0.25, 'normal': 0.125, 'bad': 0.625},
        'bad': {'good': 0.25, 'normal': 0.375, 'bad': 0.375}}  # 转移概率矩阵
    stats = ['good', 'normal', 'bad']  # 隐性状态
    obs = ["travel", "running"]  # 显性状态
    rvtb, mpath = vtb(len(obs) - 1, obs, stats, p_start, p_emit, p_trans)
    rhmm = hmm(len(obs) - 1, obs, stats, p_start, p_emit, p_trans)
    x = max(rhmm, key=rhmm.get)  # 算出概率最大的key
    print(rhmm)
    print(x)
 

 输出结果：

?

1 2

{'good-good': 0.008749999999999999, 'good-normal': 0.006562499999999999, 'good-bad': 0.0013124999999999999, 'normal-good': 0.009375, 'normal-normal': 0.0046875, 'normal-bad': 0.014062499999999999, 'bad-good': 0.0025000000000000005, 'bad-normal': 0.0037500000000000007, 'bad-bad': 0.0022500000000000003} normal-bad

算出第一天是travel，第二天是running的心情最大概率对应的心情

normal-bad，第一天心情是一般，第二天心情是糟糕的。