2

我正在展平一个数据框，其中列包含字典列表。我已经为它编写了代码。但是，仅处理 5000 行大约需要 25 秒，这非常多。

这是示例数据集：

event_date  timestamp   event_name      user_properties
20191117    1.57401E+15 user_engagement [{'key': 'ga_session_id', 'value': {'string_value': None, 'int_value': 1574005142, 'float_value': None, 'double_value': None, 'set_timestamp_micros': 1574005142713000}}, {'key': 'ga_session_number', 'value': {'string_value': None, 'int_value': 5, 'float_value': None, 'double_value': None, 'set_timestamp_micros': 1574005142713000}}, {'key': 'first_open_time', 'value': {'string_value': None, 'int_value': 1573974000000, 'float_value': None, 'double_value': None, 'set_timestamp_micros': 1573971590380000}}]
20191117    1.57401E+15 screen_view     [{'key': 'ga_session_id', 'value': {'string_value': None, 'int_value': 1574005142, 'float_value': None, 'double_value': None, 'set_timestamp_micros': 1574005142713000}}, {'key': 'ga_session_number', 'value': {'string_value': None, 'int_value': 5, 'float_value': None, 'double_value': None, 'set_timestamp_micros': 1574005142713000}}, {'key': 'first_open_time', 'value': {'string_value': None, 'int_value': 1573974000000, 'float_value': None, 'double_value': None, 'set_timestamp_micros': 1573971590380000}}]
20191117    1.57401E+15 user_engagement [{'key': 'ga_session_id', 'value': {'string_value': None, 'int_value': 1574005142, 'float_value': None, 'double_value': None, 'set_timestamp_micros': 1574005142713000}}, {'key': 'ga_session_number', 'value': {'string_value': None, 'int_value': 5, 'float_value': None, 'double_value': None, 'set_timestamp_micros': 1574005142713000}}, {'key': 'first_open_time', 'value': {'string_value': None, 'int_value': 1573974000000, 'float_value': None, 'double_value': None, 'set_timestamp_micros': 1573971590380000}}]
20191117    1.57401E+15 user_engagement [{'key': 'ga_session_id', 'value': {'string_value': None, 'int_value': 1574005142, 'float_value': None, 'double_value': None, 'set_timestamp_micros': 1574005142713000}}, {'key': 'ga_session_number', 'value': {'string_value': None, 'int_value': 5, 'float_value': None, 'double_value': None, 'set_timestamp_micros': 1574005142713000}}, {'key': 'first_open_time', 'value': {'string_value': None, 'int_value': 1573974000000, 'float_value': None, 'double_value': None, 'set_timestamp_micros': 1573971590380000}}]
20191117    1.57401E+15 user_engagement [{'key': 'ga_session_id', 'value': {'string_value': None, 'int_value': 1574005142, 'float_value': None, 'double_value': None, 'set_timestamp_micros': 1574005142713000}}, {'key': 'ga_session_number', 'value': {'string_value': None, 'int_value': 5, 'float_value': None, 'double_value': None, 'set_timestamp_micros': 1574005142713000}}, {'key': 'first_open_time', 'value': {'string_value': None, 'int_value': 1573974000000, 'float_value': None, 'double_value': None, 'set_timestamp_micros': 1573971590380000}}]

这是解析的数据框：

扁平化数据框

结果包含“键”作为列，但是，如果字典中有“set_timestamp_micros”键，则该列的语法为 {key}.set_timestamp_micros。

这是展平数据框的代码：

def normalize_complex_column_v2(df, df_copy, column):
    col_list = []
    for index,row in df.iterrows():
        for element in row[column]:
            cols = [element['key']]
            cols += ["%s.%s"%(element['key'],key) for key in element['value'].keys() if 'timestamp' in key]
            df_copy= df_copy.reindex(columns=list(dict.fromkeys(df_copy.columns.tolist() + cols)))
            df_copy.loc[index,cols] = [value for key,value in element['value'].items() if value is not None]
    df_copy.drop([column], axis=1, inplace=True)
    return df_copy

如何优化此代码？

更新：有什么方法可以使用swifter 来优化我的功能？

Numba 的问题：

<ipython-input-101-15265d3af7fb>:1: NumbaWarning: 
Compilation is falling back to object mode WITH looplifting enabled because Function "flatten_dataframe_column" failed type inference due to: Untyped global name 'defaultdict': cannot determine Numba type of <class 'type'>

File "<ipython-input-101-15265d3af7fb>", line 4:
def flatten_dataframe_column(df,column,fetch_timestamp=True):
    <source elided>
    temp_dict = df[column].to_dict()
    new_dict = defaultdict(dictLoweringError: Failed in object mode pipeline (step: object mode backend)
$22.3.182

File "<ipython-input-101-15265d3af7fb>", line 16:
def flatten_dataframe_column(df,column,fetch_timestamp=True):
    <source elided>
                        elements['key'] : [value for key,value in elements['value'].items() \
                                                    if (value is not None and 'timestamp' not in key)][0]
                                                    ^

[1] During: lowering "$22.3.182 = unary(fn=<built-in function not_>, value=$22.3.182)" at <ipython-input-101-15265d3af7fb> (16)

-------------------------------------------------------------------------------
This should not have happened, a problem has occurred in Numba's internals.
You are currently using Numba version 0.47.0.

Please report the error message and traceback, along with a minimal reproducer
at: https://github.com/numba/numba/issues/new

If more help is needed please feel free to speak to the Numba core developers
directly at: https://gitter.im/numba/numba

Thanks in advance for your help in improving Numba!

)
4

2 回答 2

0

我将数据框列转换为字典并在那里处理数据。然后将处理后的字典转换为数据框，并通过“索引”与原始数据框连接。处理 50 万条记录大约需要 8 秒。

def flatten_dataframe_column(df,column):
    temp_dict = df[column].to_dict()
    new_dict = defaultdict(dict)
    for item in temp_dict.items():
        for elements in item[1]:
               new_dict[item[0]].update(
                        {
                            (elements['key']+'.set_timestamp_micros') : elements['value']['set_timestamp_micros']
                        }
                )
                new_dict[item[0]].update(
                    { 
                        elements['key'] : [value for key,value in elements['value'].items() \
                                                    if (value is not None and 'timestamp' not in key)][0]
                    }
                )
    return pd.DataFrame.from_dict(new_dict,orient='index')

如果有人能想到更优化的解决方案，请发帖。

于 2020-01-17T13:09:27.430 回答
0

对于初学者，您可以遍历列中的值而不是整个数据框并释放一些内存。其次，您的列表理解是“循环的”。第三，复制和删除数据框的计算效率很低。

def normalize_complex_column_v2(df, df_copy, column):
    col_list = []
    for i, element in enumerate(df[column].values):
        # Get dictionary in list
        element = element[0] if type(element)==list else None
        # Optimize the code below by more efficiently looping through keys
        cols = [key for key in element.get('value').keys() if 'timestamp' in key]

        # Get values for your column:
        for key in cols:
            df.iloc[i, key] = element.get('value').get(key)

        # Now create the column for `element['key']`
        df.iloc[i, element.get('key')] = 'foo' # Some value for this, not sure where you're pulling from...

    return df

这应该可以解决问题。让我知道它是如何比较的！

于 2020-01-16T07:19:14.657 回答