nlp homework 03

来源:互联网 发布:pro recorder软件下载 编辑:程序博客网 时间:2024/06/16 09:17

NLP Homework 03 ——冯煜博

题目描述

(盒子和球模型)假设有3个盒子,每个盒子里装有红白两种颜色的球,盒子里的红白球有下表列出,初始状态分布。

符号发射概率

状态转移矩阵

解答

1. 给出HMM模型 μ=(A,B,π)

A 如下矩阵:

Abox1box2box3box10.50.30.2box20.20.50.3box30.30.20.5

B 如下矩阵

Bbox1box2box3red1225710white1235310

π 为如下矩阵

πbox10.2box20.4box30.4

2. 试用前向算法计算观测序列概率:{红,白,红}

O = {红, 白, 红}

step 1 初始化

a(box1)=πbox1bbox1()=0.2×12=0.1

a(box2)=πbox2bbox2()=0.4×25=0.16

a(box3)=πbox3bbox3()=0.4×710=0.28

step 2 归纳计算

a(box1)=[i=13a(boxi)aboxibox1]bbox1()=[a(box1)×0.5+a(box2)×0.3+a(box3)×0.2]×12=0.077

a(box2)=[i=13a(boxi)aboxibox2]bbox2()=[a(box1)×0.2+a(box2)×0.5+a(box3)×0.3]×35=0.1104

a(box3)=[i=13a(boxi)aboxibox3]bbox3()=[a(box1)×0.3+a(box2)×0.2+a(box3)×0.5]×310=0.0606

a(box1)=[i=13a(boxi)aboxibox1]bbox1()=[a(box1)×0.5+a(box2)×0.3+a(box3)×0.2]×12=0.04187

a(box2)=[i=13a(boxi)aboxibox2]bbox2()=[a(box1)×0.2+a(box2)×0.5+a(box3)×0.3]×25=0.035512

a(box3)=[i=13a(boxi)aboxibox3]bbox3()=[a(box1)×0.3+a(box2)×0.2+a(box3)×0.5]×710=0.052836

step 3 求和终结

P(O|μ)=i=13a(boxi)=a(box1)+a(box2)+a(box3)=0.04187+0.035512+0.052836=0.130218

3. 试用维特比算法,求解观测序列{红,白,红}的最优状态序列。 (要求给详细计算过程)

假设从盒子中取球是放回抽样概率模型。

step 1 初始化

δ(box1)=πbox1bbox1()=0.2×12=0.1

δ(box2)=πbox2bbox2()=0.4×25=0.16

δ(box3)=πbox3bbox3()=0.4×710=0.28

φ(box1)=0

φ(box2)=0

φ(box3)=0

step 2 归纳计算 1

δ(box1)=max[δ(box1)abox1box1,δ(box2)abox2box1,δ(box3)abox3box1]bbox1()=0.28×0.2×12=0.028

δ(box2)=max[δ(box1)abox1box2,δ(box2)abox2box2,δ(box3)abox3box2]bbox1()=0.28×0.3×35=0.0504

δ(box3)=max[δ(box1)abox1box3,δ(box2)abox2box3,δ(box3)abox3box3]bbox3()=0.28×0.5×310=0.042

记忆回退路径 1

φ(box1)=argmax[δ(box1)abox1box1,δ(box2)abox2box1,δ(box3)abox3box1]bbox1()=12δ(box3)abox3box1

φ(box2)=argmax[δ(box1)abox1box2,δ(box2)abox2box2,δ(box3)abox3box2]bbox2()=35δ(box3)abox3box2

δ(box3)=argmax[δ(box1)abox1box3,δ(box2)abox2box3,δ(box3)abox3box3]bbox3()=310δ(box3)abox3box3

归纳计算 2

δ(box1)=max[δ(box1)abox1box1,δ(box2)abox2box1,δ(box3)abox3box1]bbox1()=0.0504×0.3×12=0.0756

δ(box2)=max[δ(box1)abox1box2,δ(box2)abox2box2,δ(box3)abox3box2]bbox1()=0.0504×0.5×25=0.01008

δ(box3)=max[δ(box1)abox1box3,δ(box2)abox2box3,δ(box3)abox3box3]bbox3()=0.042×0.5×710=0.0147

记忆回退路径 2

φ(box1)=argmax[δ(box1)abox1box1,δ(box2)abox2box1,δ(box3)abox3box1]bbox1()=12δ(box2)abox2box1

φ(box2)=argmax[δ(box1)abox1box2,δ(box2)abox2box2,δ(box3)abox3box2]bbox2()=25δ(box2)abox2box2

δ(box3)=argmax[δ(box1)abox1box3,δ(box2)abox2box3,δ(box3)abox3box3]bbox3()=710δ(box3)abox3box3

step 3 终结

Q^=argmax[δ(box1),δ(box2),δ(box3)]=δ(box3)

P^(Q^)=max[δ(box1),δ(box2),δ(box3)]=0.0147

step 4 路径(状态序列)回溯

q^=φ(q^)=φ[δ(box3)]

q^=φ(q^)=φ[δ(box2)]

q^=φ(q^)=φ[δ(box3)]

0 0