gistfile1.pyw

#! -*- coding:utf-8 -*-
import random
import numpy as np


def p(str):
    print str
    pass


def p2(str):
    print str,
    pass


class HMM():
    def __init__(self):
        # 状態
        self.state = [ "A", "B", "C", "D", "E","$"]
        # 状態遷移確率
        self.trans = [ 
              [ 0, 5, 1, 3, 0, 0], #An1
              [ 1, 0, 1, 1, 1, 1], #An2
              [ 1, 0, 0, 1, 1, 1],
              [ 1, 0, 1, 1, 1, 1],
              [ 0, 0, 2, 1, 1, 0],
              [ 0, 0, 0, 0, 0, 1]
            ]
        # 出力文字
        self.outStrings = [ "天使", "ちゃん", "マジ", "！", "ユイにゃん","$"]
        self.outProbabilities = [
              [ 1, 0, 0, 0, 0, 0 ],
              [ 0, 1, 0, 0, 0, 0 ],
              [ 0, 0, 1, 0, 0, 0 ],
              [ 0, 0, 0, 1, 0, 0 ],
              [ 0, 0, 0, 0, 1, 0 ],
              [ 0, 0, 0, 0, 0, 1 ],
            ]

        self.position = self.outPosition = 0
        

    # 文字列を生成して、次の候補を探す
    def generateState(self):
        self.position = self.generate(self.trans)



    # 次の単語を状態から引っ張ってくる
    def generateWord(self):
        self.outPosition = self.generate(self.outProbabilities)


    def generate(self, prob):
        nextCandidates = prob[self.position]
        
        #print nextCandidates
        cands = []
        probs = []
        for i in range(len(nextCandidates)):
            if nextCandidates[i] >= 1:
                cands.append(i)
                probs.append(nextCandidates[i])
        probs = np.array(probs, dtype=float)
        probs /= sum(probs)
        return self.choiseNextCandidate(cands, probs)


    # 割合に応じて次の候補を選択します
    def choiseNextCandidate(self, cands, probs):
        p2(cands)
        p2(probs)
        n = np.random.multinomial(1, probs).argmax()
        p2(cands[n])
        return cands[n]

    
    def getState(self):
        return self.state[self.position]


    def getOutStrings(self):
        return self.outStrings[self.outPosition]     

    
    
from collections import defaultdict
# choiseNextCandidate がだいたい割合通りかのテスト
def testChoiseNextCandidate_Probability():
    testData = [1,2,3]
    probs = np.array([1,2, 3], dtype=float)
    probs /= sum(probs)
    result = defaultdict(int)

    for i in xrange(100):
        n = HMM().choiseNextCandidate(testData, probs)
        result[n] += 1
    p(result)


#testChoiseNextCandidate_Probability()
answer = []
a = HMM()
s = a.getState()
p("現在の状態 出力候補 出力確率 出力選択 遷移候補 遷移確率 遷移選択")
while s != '$':
    p2(s)
    a.generateWord()
    o = a.getOutStrings()
    answer.append(o)
    a.generateState()
    s = a.getState()

    p("")

p("".join(answer))