如何将带有字典列表的 pandas 列拆分为每个键的单独列

Question

我正在分析 来自 Facebook 的政治广告，这是一个 dataset released here，由 ProPublica 提供。

这就是我的意思。 我有一整列要分析的目标，但它的格式对于我的技能水平的人来说非常难以访问。

这仅来自 1 个单元格： [{"target": "NAge", "segment": "21 and older"}, {"target": "MinAge", "segment": "21"}, {"target": "Retargeting", "segment": "people who may be similar to their customers"}, {"target": "Region", "segment": "the United States"}]

还有一个： [{"target": "NAge", "segment": "18 and older"}, {"target": "Location Type", "segment": "HOME"}, {"target": "Interest", "segment": "Hispanic culture"}, {"target": "Interest", "segment": "Republican Party (United States)"}, {"target": "Location Granularity", "segment": "country"}, {"target": "Country", "segment": "the United States"}, {"target": "MinAge", "segment": 18}]

我需要做的是将每个“目标”项目分开，成为列标签，每个对应的“段”成为该列中的可能值。

或者，解决方案是创建一个函数来调用每行中的每个字典键来计算频率吗？

Answer 1

• 列为 lists，共 dicts。
  • 可以使用 pandas.explode().
  将 list 中的每个 dict 移动到单独的列
  • 将 dicts 的列转换为数据框，其中键是列 headers，值是观察值，通过使用 pandas.json_normalize()、.join() 这回到 df.
• 使用 .drop() 删除不需要的列。
• 如果该列包含字符串形式的字典列表（例如 "[{key: value}]"），请参考此 solution in Splitting dictionary/list inside a Pandas Column into Separate Columns，并使用：
  • df.col2 = df.col2.apply(literal_eval), from ast import literal_eval.

import pandas as pd

# create sample dataframe
df = pd.DataFrame({'col1': ['x', 'y'], 'col2': [[{"target": "NAge", "segment": "21 and older"}, {"target": "MinAge", "segment": "21"}, {"target": "Retargeting", "segment": "people who may be similar to their customers"}, {"target": "Region", "segment": "the United States"}], [{"target": "NAge", "segment": "18 and older"}, {"target": "Location Type", "segment": "HOME"}, {"target": "Interest", "segment": "Hispanic culture"}, {"target": "Interest", "segment": "Republican Party (United States)"}, {"target": "Location Granularity", "segment": "country"}, {"target": "Country", "segment": "the United States"}, {"target": "MinAge", "segment": 18}]]})

# display(df)
  col1                                                                                                                                                                                                                                                                                                                                                                                 col2
0    x                                                                                                                                                   [{'target': 'NAge', 'segment': '21 and older'}, {'target': 'MinAge', 'segment': '21'}, {'target': 'Retargeting', 'segment': 'people who may be similar to their customers'}, {'target': 'Region', 'segment': 'the United States'}]
1    y  [{'target': 'NAge', 'segment': '18 and older'}, {'target': 'Location Type', 'segment': 'HOME'}, {'target': 'Interest', 'segment': 'Hispanic culture'}, {'target': 'Interest', 'segment': 'Republican Party (United States)'}, {'target': 'Location Granularity', 'segment': 'country'}, {'target': 'Country', 'segment': 'the United States'}, {'target': 'MinAge', 'segment': 18}]

# use explode to give each dict in a list a separate row
df = df.explode('col2').reset_index(drop=True)

# normalize the column of dicts, join back to the remaining dataframe columns, and drop the unneeded column
df = df.join(pd.json_normalize(df.col2)).drop(columns=['col2'])

display(df)

   col1                target                                       segment
0     x                  NAge                                  21 and older
1     x                MinAge                                            21
2     x           Retargeting  people who may be similar to their customers
3     x                Region                             the United States
4     y                  NAge                                  18 and older
5     y         Location Type                                          HOME
6     y              Interest                              Hispanic culture
7     y              Interest              Republican Party (United States)
8     y  Location Granularity                                       country
9     y               Country                             the United States
10    y                MinAge                                            18

得到count

• 如果目标是为每个 'target' 和关联的 'segment'
获取 count

counts = df.groupby(['target', 'segment']).count()

已更新

• 此更新是针对完整文件实施的

import pandas as pd
from ast import literal_eval

# load the file
df = pd.read_csv('en-US.csv')

# replace NaNs with '[]', otherwise literal_eval will error
df.targets = df.targets.fillna('[]')

# replace null with None, otherwise literal_eval will error
df.targets = df.targets.str.replace('null', 'None')

# convert the strings to lists of dicts
df.targets = df.targets.apply(literal_eval)

# use explode to give each dict in a list a separate row
df = df.explode('targets').reset_index(drop=True)

# fillna with {} is required for json_normalize
df.targets = df.targets.fillna({i: {} for i in df.index})

# normalize the column of dicts, join back to the remaining dataframe columns, and drop the unneeded column
normalized = pd.json_normalize(df.targets)

# get the counts
counts = normalized.groupby(['target', 'segment']).segment.count().reset_index(name='counts')