简化 - 如果混乱更好的结构?

时间:2011-01-12 15:58:06

标签: python markov-chains

请将此问题移至Code Review -area。它更适合那里,因为我知道下面的代码是垃圾,我希望关键反馈完成重写。我正在重新发明轮子。

# Description: you are given a bitwise pattern and a string
# you need to find the number of times the pattern matches in the string.
# The pattern is determined by markov chain.
# For simplicity, suppose the ones and zeros as unbiased coin flipping
# that stops as it hits the pattern, below.
#
# Any one liner or simple pythonic solution?

import random

def matchIt(yourString, yourPattern):
        """find the number of times yourPattern occurs in yourString"""

        count = 0
        matchTimes = 0

        # How can you simplify the for-if structures?
        # THIS IS AN EXAMPLE HOW NOT TO DO IT, hence Code-Smell-label
        # please, read clarifications in [Update]

        for coin in yourString:
            #return to base
            if  count == len(pattern):
                    matchTimes = matchTimes + 1
                    count = 0

            #special case to return to 2, there could be more this type of conditions
            #so this type of if-conditionals are screaming for a havoc
            if count == 2 and pattern[count] == 1:
                    count = count - 1

            #the work horse
            #it could be simpler by breaking the intial string of lenght 'l'
            #to blocks of pattern-length, the number of them is 'l - len(pattern)-1'
            if coin == pattern[count]:
                    count=count+1

        average = len(yourString)/matchTimes

        return [average, matchTimes]



# Generates the list
myString =[]
for x in range(10000):
    myString= myString + [int(random.random()*2)]

pattern = [1,0,0]
result = matchIt(myString, pattern)

print("The sample had "+str(result[1])+" matches and its size was "+str(len(myString))+".\n" +
        "So it took "+str(result[0])+" steps in average.\n" +
        "RESULT: "+str([a for a in "FAILURE" if result[0] != 8]))


# Sample Output
# 
# The sample had 1656 matches and its size was 10000.
# So it took 6 steps in average.
# RESULT: ['F', 'A', 'I', 'L', 'U', 'R', 'E']

[更新]

我将在这里解释一下理论,或许,问题可以通过这种方式简化。上面的代码尝试使用下面的转换矩阵A构造马尔可夫链。您可以想象为硬币翻转的模式100对应于它。

>>> Q=numpy.matrix('0.5 0.5 0; 0 0.5 0.5; 0 0.5 0')     
>>> I=numpy.identity(3)
>>> I
array([[ 1.,  0.,  0.],
       [ 0.,  1.,  0.],
       [ 0.,  0.,  1.]])
>>> Q
matrix([[ 0.5,  0.5,  0. ],
        [ 0. ,  0.5,  0.5],
        [ 0. ,  0.5,  0. ]])
>>> A=numpy.matrix('0.5 0.5 0 0; 0 0.5 0.5 0; 0 0.5 0 0.5; 0 0 0 1')
>>> A
matrix([[ 0.5,  0.5,  0. ,  0. ],
        [ 0. ,  0.5,  0.5,  0. ],
        [ 0. ,  0.5,  0. ,  0.5],
        [ 0. ,  0. ,  0. ,  1. ]])

问题中的average 8成为矩阵N=(I-Q)^-1中第一行的值的总和,其中Q位于上方。

>>> (I-Q)**-1
matrix([[ 2.,  4.,  2.],
        [ 0.,  4.,  2.],
        [ 0.,  2.,  2.]])
>>> numpy.sum(((I-Q)**-1)[0])
8.0

现在,你可能会发现这个明显只有模式匹配的问题变成了马尔可夫链。我无法理解为什么你不能用类似于矩阵或矩阵的东西替换凌乱的for-while-if条件。我不知道如何实现它们,但迭代器可能是一种方法,研究,特别是在需要分解的更多状态。

但是Numpy出现了一个问题,-InfNaN的内容是什么?从(I-Q)**-1矩阵检查上面应该收敛的值。 N来自N=I+Q+Q^2+Q^3+...=\frac{I-Q^{n}}{I-Q}

>>> (I-Q**99)/(I-Q)
matrix([[  2.00000000e+00,   1.80853571e-09,             -Inf],
        [             NaN,   2.00000000e+00,   6.90799171e-10],
        [             NaN,   6.90799171e-10,   1.00000000e+00]])
>>> (I-Q**10)/(I-Q)
matrix([[ 1.99804688,  0.27929688,        -Inf],
        [        NaN,  1.82617188,  0.10742188],
        [        NaN,  0.10742188,  0.96679688]])

3 个答案:

答案 0 :(得分:2)

def matchIt(yourString, yourPattern):
        """find the number of times yourPattern occurs in yourString"""

您是否可以使用以下内容?

yourString.count(yourPattern)

在您的情况下,您可以将myString创建为10000字符的真实字符串,将pattern也创建为字符串,然后以简单的pythonic方式计算出现的次数。

修改

pattern(可以是字符串或列表)中为您提供text的(重叠)出现次数的单行,可能如下所示:

nbOccurences = sum(1 for i in xrange(len(text)-len(pattern)) if text[i:i+len(pattern)] == pattern)

答案 1 :(得分:1)

好的 - 标准(-ish)字符串搜索:

def matchIt(needle, haystack):
    """
    @param needle:   string, text to seek
    @param haystack: string, text to search in

    Return number of times needle is found in haystack,
        allowing overlapping instances.

    Example: matchIt('abab','ababababab') -> 4
    """
    lastSeenAt = -1
    timesSeen = 0
    while True:
        nextSeen = haystack.find(needle, lastSeenAt+1)
        if nextSeen==-1:
            return timesSeen
        else:
            lastSeenAt = nextSeen
            timesSeen += 1

但你想这样做到一个数字列表?没问题;我们只需要使用find()方法创建一个列表类,如下所示:

import itertools
class FindableList(list):
    def find(self, sub, start=None, end=None):
        """
        @param sub: list, pattern to look for in self

        @param start: int, first possible start-of-list
            If not specified, start at first item

        @param: end: int, last+1 possible start-of-list
            If not specified, end such that entire self is searched

        Returns;
            Starting offset if a match is found, else -1
        """
        if start is None or start < 0:
            start = 0

        # N.B. If end is allowed to be too high,
        # zip() will silently truncate the list comparison
        # and you will probably get extra spurious matches.
        lastEnd = len(self) - len(sub) + 1
        if end is None or end > lastEnd:
            end = lastEnd

        rng = xrange if xrange else range
        iz  = itertools.izip
        isl = itertools.islice

        for pos in rng(start, end):
            if all(a==b for a,b in iz(sub, isl(self, pos, end))):
                return pos

        # no match found
        return -1

然后该示例看起来像

matchIt([1,2,1,2], FindableList([1,2,1,2,1,2,1,2,1,2])) -> 4

,您的代码变为:

# Generate a list
randIn = lambda x: int(x*random.random())
myString =[randIn(2) for i in range(10000)]

pattern = [1,0,0]
result = matchIt(pattern, myString)

print("The sample had {0} matches and its size was {1}.\n".format(result, len(myString)))

答案 2 :(得分:0)

尚未就绪。

类似的问题,但主要关注图形库here和类似的问题,但在C#,可能有用。

与此问题相关的文件是./networkx/generators/degree_seq.py(997行,关于生成具有给定度数序列的graps)和./networkx/algorithms/mixing.py (line 20, function degree_assortativity(G) about probability based graphs)并且还注意其源代码引用92个引用,不确定是否你想重新发明轮子。对于igraph,请阅读文件convert.c关于加权边的第835行。您可以获取Networkx here的来源和igraph here的来源。请注意,前者是BSD许可证,用Python完成,而igraph在GNU(GPL)下完成,用C语言完成。

要开始使用Networkx,有关从jUnits test_convert_scipy.py文件创建加权图的有用信息:

def create_weighted(self, G): 
    g = cycle_graph(4)
    e = g.edges()
    source = [u for u,v in e]
    dest = [v for u,v in e]
    weight = [s+10 for s in source]
    ex = zip(source, dest, weight)
    G.add_weighted_edges_from(ex)
    return G

所以要创建你的马尔可夫链,请提供有关定向加权图here的帮助,或许这样:

>>> DG=nx.DiGraph()
>>> DG.add_weighted_edges_from([(0,0,0.5),(1,1,0.5),(3,3,1),(0,1,0.5),(1,2,0.5),(2,3,0.5), (2,1,0.5)])

或者可能存在一些现成的马尔可夫链生成工具,因为有一些其他随机过程,更多here。无法找到algorithm来分析具有例外值的图表,或者在示例中使用不同的集合进行试验,也许没有,并且您必须坚持使用其他回复者的解决方案。