python - 从 json 对象创建 pandas 数据框

Question

我终于从一个包含许多 json 对象的文件中得到了我需要的数据输出，但是我需要一些帮助来将下面的输出转换为单个数据帧，因为它循环遍历数据。以下是生成输出的代码，包括输出的示例:

原始数据:

{
"zipcode":"08989",
"current"{"canwc":null,"cig":4900,"class":"observation","clds":"OVC","day_ind":"D","dewpt":19,"expireTimeGMT":1385486700,"feels_like":34,"gust":null,"hi":37,"humidex":null,"icon_code":26,"icon_extd":2600,"max_temp":37,"wxMan":"wx1111"},
"triggers":[53,31,9,21,48,7,40,178,55,179,176,26,103,175,33,51,20,57,112,30,50,113]
}
{
"zipcode":"08990",
"current":{"canwc":null,"cig":4900,"class":"observation","clds":"OVC","day_ind":"D","dewpt":19,"expireTimeGMT":1385486700,"feels_like":34,"gust":null,"hi":37,"humidex":null,"icon_code":26,"icon_extd":2600,"max_temp":37, "wxMan":"wx1111"},
"triggers":[53,31,9,21,48,7,40,178,55,179,176,26,103,175,33,51,20,57,112,30,50,113]
}

def lines_per_n(f, n):
    for line in f:
        yield ''.join(chain([line], itertools.islice(f, n - 1)))

for fin in glob.glob('*.txt'):
    with open(fin) as f:
        for chunk in lines_per_n(f, 5):
            try:
                jfile = json.loads(chunk)
                zipcode = jfile['zipcode']
                datetime = jfile['current']['proc_time']
                triggers = jfile['triggers']
                print pd.Series(jfile['zipcode']), 
                      pd.Series(jfile['current']['proc_time']),\
                      jfile['triggers']          
            except ValueError, e:
                pass
            else:
                pass

我在运行上面的代码时得到的示例输出，我想将其作为 3 列存储在 pandas 数据框中。

08988 20131126102946 []
08989 20131126102946 [53, 31, 9, 21, 48, 7, 40, 178, 55, 179]
08988 20131126102946 []
08989 20131126102946 [53, 31, 9, 21, 48, 7, 40, 178, 55, 179]
00544 20131126102946 [178, 30, 176, 103, 179, 112, 21, 20, 48]

所以下面的代码看起来更接近，因为如果我在列表中传递 df 并转置 df，它会给我一个时髦的 df。关于如何正确 reshape 它的任何想法？

def series_chunk(chunk):
    jfile = json.loads(chunk)
    zipcode = jfile['zipcode']
    datetime = jfile['current']['proc_time']
    triggers = jfile['triggers']
    return jfile['zipcode'],\
            jfile['current']['proc_time'],\
            jfile['triggers']

for fin in glob.glob('*.txt'):
    with open(fin) as f:
        for chunk in lines_per_n(f, 7):
            df1 = pd.DataFrame(list(series_chunk(chunk)))
            print df1.T

[u'08988', u'20131126102946', []]
[u'08989', u'20131126102946', [53, 31, 9, 21, 48, 7, 40, 178, 55, 179]]
[u'08988', u'20131126102946', []]
[u'08989', u'20131126102946', [53, 31, 9, 21, 48, 7, 40, 178, 55, 179]]

数据框:

   0               1   2
0  08988  20131126102946  []
       0               1                                                  2
0  08989  20131126102946  [53, 31, 9, 21, 48, 7, 40, 178, 55, 179, 176, ...
       0               1   2
0  08988  20131126102946  []
       0               1                                                  2
0  08989  20131126102946  [53, 31, 9, 21, 48, 7, 40, 178, 55, 179, 176, ...

这是我的最终代码和输出。我如何捕获它通过循环创建的每个数据帧并将它们动态连接为一个数据帧对象？

for fin in glob.glob('*.txt'):
    with open(fin) as f:
        print pd.concat([series_chunk(chunk) for chunk in lines_per_n(f, 7)], axis=1).T

       0               1                                                  2
0  08988  20131126102946                                                 []
1  08989  20131126102946  [53, 31, 9, 21, 48, 7, 40, 178, 55, 179, 176, ...
       0               1                                                  2
0  08988  20131126102946                                                 []
1  08989  20131126102946  [53, 31, 9, 21, 48, 7, 40, 178, 55, 179, 176, ...

Answer 1

###注意:对于那些想将 json 解析为 pandas 的人，如果你有 valid json(这个问题没有)，那么你应该使用 pandas read_json 函数:

# can either pass string of the json, or a filepath to a file with valid json
In [99]: pd.read_json('[{"A": 1, "B": 2}, {"A": 3, "B": 4}]')
Out[99]:
   A  B
0  1  2
1  3  4

查看 IO part of the docs 中的几个示例、可以传递给此函数的参数，以及规范化较少结构化 json 的方法。

如果您没有有效的 json，在读入 json 之前对字符串进行处理通常是有效的，例如 see this answer 。

如果您有多个 json 文件，您应该将 DataFrame 连接在一起(类似于此答案):

pd.concat([pd.read_json(file) for file in ...], ignore_index=True)

###这个例子的原始答案:

在传递给 read_csv 的分隔符的正则表达式中使用 lookbehind:

In [11]: df = pd.read_csv('foo.csv', sep='(?<!,)\s', header=None)

In [12]: df
Out[12]: 
       0               1                                                  2
0   8988  20131126102946                                                 []
1   8989  20131126102946  [53, 31, 9, 21, 48, 7, 40, 178, 55, 179, 176, ...
2   8988  20131126102946                                                 []
3   8989  20131126102946  [53, 31, 9, 21, 48, 7, 40, 178, 55, 179, 176, ...
4    544  20131126102946  [178, 30, 176, 103, 179, 112, 21, 20, 48, 7, 5...
5    601  20131126094911                                                 []
6    602  20131126101056                                                 []
7    603  20131126101056                                                 []
8    604  20131126101056                                                 []
9    544  20131126102946  [178, 30, 176, 103, 179, 112, 21, 20, 48, 7, 5...
10   601  20131126094911                                                 []
11   602  20131126101056                                                 []
12   603  20131126101056                                                 []
13   604  20131126101056                                                 []

[14 rows x 3 columns]

如评论中所述，您可以通过将多个系列连接在一起来更直接地做到这一点......它也将更容易理解:

def series_chunk(chunk):
    jfile = json.loads(chunk)
    zipcode = jfile['zipcode']
    datetime = jfile['current']['proc_time']
    triggers = jfile['triggers']
    return pd.Series([jfile['zipcode'], jfile['current']['proc_time'], jfile['triggers']])

dfs = []
for fin in glob.glob('*.txt'):
    with open(fin) as f:
        df = pd.concat([series_chunk(chunk) for chunk in lines_per_n(f, 5)], axis=1)
        dfs.append(dfs)

df = pd.concat(dfs, ignore_index=True)

注意:您也可以将 try/except 移动到 series_chunk 中。