Jiebe学习笔记：viterbi算法

    词性标注:

    Jieba的词性标注模块仍然采用基于HMM模型的viterbi算法，在状态的描述上应用（{BMES}，词性）二元对。状态集合为{BMSE}这四个词位与词性（39类）的笛卡尔积，也就是有156种状态。

    从文本的起始位置开始逐字扫描，判断每个字可能的状态，与状态转移得到的期待状态求与，也就是下一个字可能的状态。与的结果是空集时以期待状态为下一个字的状态，期待状态集为空时，以整个状态集为其状态。

    

def viterbi(obs, states, start_p, trans_p, emit_p):#obs:待处理文本，states，文字可能的状态集，start_p:初始概率，
    V = [{}]  # tabular
    mem_path = [{}]
    all_states = trans_p.keys()
    for y in states.get(obs[0], all_states):  # init
    #get返回turple，表示首字所有可能的状态
        V[0][y] = start_p[y] + emit_p[y].get(obs[0], MIN_FLOAT)
        mem_path[0][y] = ''
    for t in xrange(1, len(obs)):
        V.append({})
        mem_path.append({})
        #prev_states = get_top_states(V[t-1])
        prev_states = [
            x for x in mem_path[t - 1].keys() if len(trans_p[x]) > 0]
        #如果下一状态集不为空，x为上一状态
        prev_states_expect_next = set(
            (y for x in prev_states for y in trans_p[x].keys()))
        #y:下一步可能的状态
        obs_states = set(
            states.get(obs[t], all_states)) & prev_states_expect_next
        #汉字可能的状态与期待状态的交
        if not obs_states:
            obs_states = prev_states_expect_next if prev_states_expect_next else all_states

        for y in obs_states:
            prob, state = max((V[t - 1][y0] + trans_p[y0].get(y, MIN_INF) +
                               emit_p[y].get(obs[t], MIN_FLOAT), y0) for y0 in prev_states)
            V[t][y] = prob
            mem_path[t][y] = state

    last = [(V[-1][y], y) for y in mem_path[-1].keys()]
    # if len(last)==0:
    #     print obs
    prob, state = max(last)

    route = [None] * len(obs)
    i = len(obs) - 1
    while i >= 0:
        route[i] = state
        state = mem_path[i][state]
        i -= 1
    return (prob, route)