计算python中2个字典中值之间的平均绝对百分比误差

时间:2016-06-13 18:45:11

标签: python for-loop dictionary

我有一个位置字典,然后是属性 - 值对,如下所示:

{"Russia": 
    {"/location/statistical_region/size_of_armed_forces": 65700.0,
     "/location/statistical_region/gni_per_capita_in_ppp_dollars": 42530.0, 
     "/location/statistical_region/gdp_nominal": 1736050505050.0,
     "/location/statistical_region/foreign_direct_investment_net_inflows": 8683048195.0, 
     "/location/statistical_region/life_expectancy": 80.929, ...

依旧,对于每个国家。

然后是一个包含单个数组的字典,数组中的每个值都是一个包含3个键的字典:

{
    "sentences": [
        {
            "location-value-pair": {
                "Russia": 6.1
            }, 
            "parsedSentence": "On Tuesday , the Federal State Statistics Service -LRB- Rosstat -RRB- reported that consumer price inflation in LOCATION_SLOT hit a historic post-Soviet period low of NUMBER_SLOT percent in 2011 , citing final data .", 
            "sentence": "On Tuesday , the Federal State Statistics Service -LRB- Rosstat -RRB- reported that consumer price inflation in Russia hit a historic post-Soviet period low of 6.1 percent in 2011 , citing final data ."
        }, 
        {
            "location-value-pair": {
                "Russia": 8.8
            }, 
            "parsedSentence": "In 2010 , annual inflation in LOCATION_SLOT hit NUMBER_SLOT percent due to the summer drought , exceeding forecasts and equalling the figure for 2009 , the year of the global financial meltdown .", 
            "sentence": "In 2010 , annual inflation in Russia hit 8.8 percent due to the summer drought , exceeding forecasts and equalling the figure for 2009 , the year of the global financial meltdown ."
        }, ...

我想要做的是比较每个句子,并为该句子中的每个位置和值,计算与第一个字典中的位置 - 值对匹配的最接近的匹配值,然后返回它对应的顶级统计属性并将其添加为句子词典的新键。

例如:

对于第1句,我看到我们正在看俄罗斯,其值为6.1。我想索引到第一个字典,找到"俄罗斯",并浏览所有存在的值,例如65700.0,42530.0,1736050505050.0,8683048195.0。然后我想找到每个属性的平均绝对误差,例如size_of_armed_forces值为23%,gni_per_capita属性为10%等。然后,我想找到最小的一个,并将其作为第二个字典的键添加,所以:

{
                "location-value-pair": {
                    "Russia": 6.1
                }, 
                "predictedRegion": "/location/statistical_region/gni_in_ppp_dollars"
                "meanabserror": 2%
                "parsedSentence": "On Tuesday , the Federal State Statistics Service -LRB- Rosstat -RRB- reported that consumer price inflation in LOCATION_SLOT hit a historic post-Soviet period low of NUMBER_SLOT percent in 2011 , citing final data .", 
                "sentence": "On Tuesday , the Federal State Statistics Service -LRB- Rosstat -RRB- reported that consumer price inflation in Russia hit a historic post-Soviet period low of 6.1 percent in 2011 , citing final data ."
            }, 

在考虑编写本文时,我的困惑就是如何访问另一个字典的键值作为另一个字典的条件。我目前的想法是:

def predictRegion(sentenceArray,trueDict):

    absPercentageErrors = {}

    for location, property2value in trueDict.items():
        print location
        absPercentageErrors['location'] = {}
        for property,trueValue in property2value.iteritems():
            print property
            absError = abs(sentenceArray['sentences']['location-value-pair'].key() - trueValue)
            absPercentageErrors['location']['property'] = absError/numpy.abs(trueValue)

    for index, dataTriples in enumerate(sentenceArray["sentences"]):
        for location, trueValue in dataTriples['location-value-pair'].items():
            print location

但显然我无法访问此行中的sentenceArray['sentences']['location-value-pair'].key()absError = abs(sentenceArray['sentences']['location-value-pair'].key() - trueValue)因为它位于循环之外。

如何从一个引用完全不同的变量的循环中访问此键?

1 个答案:

答案 0 :(得分:1)

将来阅读如何制定一个好问题:https://stackoverflow.com/help/mcve

最低,完整且可验证。

我认为这就是你要找的东西。

countries = {'Canada': {'a': 10, 'b': 150, 'c': 1000},
             'Russia': {'d': 9, 'e': 5, 'f': 1e5}}
sentences = [
        {"location-value-pair": {"Russia": 6.1}, 
         "parsedSentence": "bob loblaw", 
         "sentence": "lobs law bomb"
        }, 
        {"location-value-pair": {"Russia": 8.8}, 
            "parsedSentence": "some sentence", 
            "sentence": "lorem ipsum test"
        }]


def absError(numer,denom):
    return abs(numer-denom)/float(denom)

def findMatch(target, country):
    return min(country, key= lambda x: absError(target, country.get(x)))

def update(sentence):
    (c,target), = sentence.get("location-value-pair").items()
    country = countries[c]
    matched = findMatch(target,country)
    error = absError(target, country.get(matched))
    res = sentence.copy()
    res.update({'predictedRegion': matched, 'meanabserror': "{:.2f}%".format(100*error)})
    return res

updated = [update(sentence) for sentence in sentences]    
updated 

输出:

[{'location-value-pair': {'Russia': 6.1},
  'meanabserror': '22.00%',
  'parsedSentence': 'bob loblaw',
  'predictedRegion': 'e',
  'sentence': 'lobs law bomb'},
 {'location-value-pair': {'Russia': 8.8},
  'meanabserror': '2.22%',
  'parsedSentence': 'some sentence',
  'predictedRegion': 'd',
  'sentence': 'lorem ipsum test'}]