How to pull a key from a dict (pandas series) to its own row?-CodePudding

Here is my example data with two fields where the last one [outbreak] is a pandas series.

Start:

Goal (Excel mock-up):

Reproduction Code:

import pandas as pd
import json

d = {'report_id': [100, 101], 'outbreak': [
    '{"outbreak_100":{"name":"Chris","disease":"A-Pox"},"outbreak_101":{"name":"Stacy","disease": "H-Pox"}}', 
    '{"outbreak_200":{"name":"Brandon","disease":"C-Pox"},"outbreak_201":{"name":"Karen","disease": "G-Pox"},"outbreak_202":{"name":"Tim","disease": "Z-Pox"}}']}

df = pd.DataFrame(data=d)
print(type(df['outbreak']))
display(df)

#Ignore
df = pd.json_normalize(df['outbreak'].apply(json.loads), max_level=0)
display(df)

Attempts: I thought about using json_normalize() which would convert every [outbreak_id] to its own field and then use pandas.wide_to_long() to get my final output. It works in testing but my concern is that my actual production data is so long and nested that it ends up generating hundred of thousands of fields before pivoting. That does not sounds good to me and why I also hope to avoid loop iterations.

I also thought about using df = df.explode('outbreak') but I am getting a KeyError: 0

Perhaps someone has a better idea than I do? Thank you.

CodePudding user response：

You can try with ast convert to dict format , then we do conversion

import ast 
out = df.pop('outbreak').map(ast.literal_eval).apply(pd.Series).stack().reset_index(level=1).join(df)
out.columns = ['outbreak_id','outbreak_value','report_id']
Out[157]: 
        level_1                                        0  report_id
0  outbreak_100    {'name': 'Chris', 'disease': 'A-Pox'}        100
0  outbreak_101    {'name': 'Stacy', 'disease': 'H-Pox'}        100
1  outbreak_200  {'name': 'Brandon', 'disease': 'C-Pox'}        101
1  outbreak_201    {'name': 'Karen', 'disease': 'G-Pox'}        101
1  outbreak_202      {'name': 'Tim', 'disease': 'Z-Pox'}        101

CodePudding user response：

Try this

import json
d = {'report_id': [100, 101], 'outbreak': [
    '{"outbreak_100":{"name":"Chris","disease":"A-Pox"},"outbreak_101":{"name":"Stacy","disease": "H-Pox"}}', 
    '{"outbreak_200":{"name":"Brandon","disease":"C-Pox"},"outbreak_201":{"name":"Karen","disease": "G-Pox"},"outbreak_202":{"name":"Tim","disease": "Z-Pox"}}']}

df = pd.DataFrame(data=d)
# use json.loads to parse the json and construct df from it
df = pd.DataFrame(df.set_index('report_id')['outbreak'].map(json.loads).to_dict()).stack().rename_axis(['outbreak_id', 'report_id'], axis=0).reset_index(name='outbreak_value')
print(df)
    outbreak_id  report_id                           outbreak_value
0  outbreak_100        100    {'name': 'Chris', 'disease': 'A-Pox'}
1  outbreak_101        100    {'name': 'Stacy', 'disease': 'H-Pox'}
2  outbreak_200        101  {'name': 'Brandon', 'disease': 'C-Pox'}
3  outbreak_201        101    {'name': 'Karen', 'disease': 'G-Pox'}
4  outbreak_202        101      {'name': 'Tim', 'disease': 'Z-Pox'}

CodePudding user response：

One way to do this is to convert the json for each outbreak into a dictionary, make a list of all the dictionary key/value pairs and then explode that list and convert the values into the two desired columns:

df['outbreak'] = df['outbreak'].apply(lambda v:json.loads(v).items())
df = df.explode('outbreak')
df[['outbreak_id', 'outbreak_value']] = pd.DataFrame(df.pop('outbreak').tolist(), index=df.index)

Output (for your sample data):

   report_id   outbreak_id                           outbreak_value
0        100  outbreak_100    {'name': 'Chris', 'disease': 'A-Pox'}
0        100  outbreak_101    {'name': 'Stacy', 'disease': 'H-Pox'}
1        101  outbreak_200  {'name': 'Brandon', 'disease': 'C-Pox'}
1        101  outbreak_201    {'name': 'Karen', 'disease': 'G-Pox'}
1        101  outbreak_202      {'name': 'Tim', 'disease': 'Z-Pox'}