[FIXED] How to flatten a list of dicts from a Pandas DataFrame into several columns?

Issue

I have a pandas dataframe that looks like this:

User | Query|                                 Filters                 
----------------------------------------------------------------------------------------- 
1    |  abc | [{u'Op': u'and', u'Type': u'date', u'Val': u'1992'},{u'Op': u'and', u'Type': u'sex', u'Val': u'F'}]
1    |  efg | [{u'Op': u'and', u'Type': u'date', u'Val': u'2000'},{u'Op': u'and', u'Type': u'col', u'Val': u'Blue'}] 
1    |  fgs | [{u'Op': u'and', u'Type': u'date', u'Val': u'2001'},{u'Op': u'and', u'Type': u'col', u'Val': u'Red'}]        
2    |  hij | [{u'Op': u'and', u'Type': u'date', u'Val': u'2002'}]  
2    |  dcv | [{u'Op': u'and', u'Type': u'date', u'Val': u'2001'},{u'Op': u'and', u'Type': u'sex', u'Val': u'F'}]     
2    |  tyu | [{u'Op': u'and', u'Type': u'date', u'Val': u'1999'},{u'Op': u'and', u'Type': u'col', u'Val': u'Yellow'}]     
3    |  jhg | [{u'Op': u'and', u'Type': u'date', u'Val': u'2001'},{u'Op': u'and', u'Type': u'sex', u'Val': u'M'}]    
4    |  mlh | [{u'Op': u'and', u'Type': u'date', u'Val': u'2001'}]  

The result that I expect:

User| Query |  date | sex | col
-------------------------------- 
1   | abc   | 1992  |  F  |
1   | efg   | 2000  |     | Blue
1   | fgs   | 2001  |     | Red
2   | hij   | 2002  |     |
2   | dcv   | 2001  |  F  |
2   | tyu   | 1999  |     | Yellow
3   | jhg   | 2001  |     |
4   | mlh   | 2001  |  H  |

I'm using pandas 0.21.0 with python 2.7.

Example data:

df = pd.DataFrame([{'user': 1,'query': 'abc', 'Filters': [{u'Op': u'and', u'Type': u'date', u'Val': u'1992'},{u'Op': u'and', u'Type': u'sex', u'Val': u'F'}]},
              {'user': 1,'query': 'efg', 'Filters': [{u'Op': u'and', u'Type': u'date', u'Val': u'2000'},{u'Op': u'and', u'Type': u'col', u'Val': u'Blue'}]},
              {'user': 1,'query': 'fgs', 'Filters': [{u'Op': u'and', u'Type': u'date', u'Val': u'2001'},{u'Op': u'and', u'Type': u'col', u'Val': u'Red'}]},
              {'user': 2 ,'query': 'hij', 'Filters': [{u'Op': u'and', u'Type': u'date', u'Val': u'2002'}]},
              {'user': 2 ,'query': 'dcv', 'Filters': [{u'Op': u'and', u'Type': u'date', u'Val': u'2001'},{u'Op': u'and', u'Type': u'sex', u'Val': u'F'}]},
              {'user': 2 ,'query': 'tyu', 'Filters':[{u'Op': u'and', u'Type': u'date', u'Val': u'1999'},{u'Op': u'and', u'Type': u'col', u'Val': u'Yellow'}]},
              {'user': 3 ,'query': 'jhg', 'Filters':[{u'Op': u'and', u'Type': u'date', u'Val': u'2001'},{u'Op': u'and', u'Type': u'sex', u'Val': u'M'}]},
              {'user': 4 ,'query': 'mlh', 'Filters':[{u'Op': u'and', u'Type': u'date', u'Val': u'2001'}]},
             ])

I tried many solutions:

• Handling of nested JSON records
• Pandas read nested json

Any suggestions would be much appreciated!

Solution

Assuming you have already imported your data, as defined in your MCWE:

data = [{'user': 1,'query': 'abc', 'Filters': [{u'Op': u'and', u'Type': u'date', u'Val': u'1992'},{u'Op': u'and', u'Type': u'sex', u'Val': u'F'}]},
              {'user': 1,'query': 'efg', 'Filters': [{u'Op': u'and', u'Type': u'date', u'Val': u'2000'},{u'Op': u'and', u'Type': u'col', u'Val': u'Blue'}]},
              {'user': 1,'query': 'fgs', 'Filters': [{u'Op': u'and', u'Type': u'date', u'Val': u'2001'},{u'Op': u'and', u'Type': u'col', u'Val': u'Red'}]},
              {'user': 2 ,'query': 'hij', 'Filters': [{u'Op': u'and', u'Type': u'date', u'Val': u'2002'}]},
              {'user': 2 ,'query': 'dcv', 'Filters': [{u'Op': u'and', u'Type': u'date', u'Val': u'2001'},{u'Op': u'and', u'Type': u'sex', u'Val': u'F'}]},
              {'user': 2 ,'query': 'tyu', 'Filters':[{u'Op': u'and', u'Type': u'date', u'Val': u'1999'},{u'Op': u'and', u'Type': u'col', u'Val': u'Yellow'}]},
              {'user': 3 ,'query': 'jhg', 'Filters':[{u'Op': u'and', u'Type': u'date', u'Val': u'2001'},{u'Op': u'and', u'Type': u'sex', u'Val': u'M'}]},
              {'user': 4 ,'query': 'mlh', 'Filters':[{u'Op': u'and', u'Type': u'date', u'Val': u'2001'}]},
             ]

Then, you are looking for Pandas json_normalize method for data normalization:

from pandas.io.json import json_normalize
df = json_normalize(data, 'Filters', ['query', 'user'])

It returns a normalized DataFrame version where your column of json is expanded into eponymous typed columns:

     Op  Type     Val  user query
0   and  date    1992     1   abc
1   and   sex       F     1   abc
2   and  date    2000     1   efg
3   and   col    Blue     1   efg
4   and  date    2001     1   fgs
5   and   col     Red     1   fgs
6   and  date    2002     2   hij
7   and  date    2001     2   dcv
8   and   sex       F     2   dcv
9   and  date    1999     2   tyu
10  and   col  Yellow     2   tyu
11  and  date    2001     3   jhg
12  and   sex       M     3   jhg
13  and  date    2001     4   mlh

Now, you would pivot your DataFrame to convert Type modalities into columns:

df = df.pivot_table(index=['user', 'query', 'Op'], columns='Type', aggfunc='first')

It leads to:

                   Val            
Type               col  date   sex
user query Op                     
1    abc   and    None  1992     F
     efg   and    Blue  2000  None
     fgs   and     Red  2001  None
2    dcv   and    None  2001     F
     hij   and    None  2002  None
     tyu   and  Yellow  1999  None
3    jhg   and    None  2001     M
4    mlh   and    None  2001  None

Finally, you can clean and reset index, if they bother you:

df.columns = df.columns.droplevel(0)
df.reset_index(inplace=True)

Which returns your requested MCVE output:

Type  user query   Op     col  date   sex
0        1   abc  and    None  1992     F
1        1   efg  and    Blue  2000  None
2        1   fgs  and     Red  2001  None
3        2   dcv  and    None  2001     F
4        2   hij  and    None  2002  None
5        2   tyu  and  Yellow  1999  None
6        3   jhg  and    None  2001     M
7        4   mlh  and    None  2001  None

Not column

In this final DataFrame the first column seems to be called Type, but it is not. It is instead a Integer Index without Name:

df.index
RangeIndex(start=0, stop=8, step=1)

And Columns index is called Type which does not hold any modality called Type (therefore no column with this name).

df.columns
Index(['user', 'query', 'Op', 'col', 'date', 'sex'], dtype='object', name='Type')

This is why you cannot remove the column Type (column used in pivot_table), because it does not exist.

If you want to remove this fake column, you need to create a new index for rows:

df.set_index(['user', 'query'], inplace=True)

If Column index Name bothers you, you can reset it:

df.columns.name = None

It leads to:

             Op     col  date   sex
user query                         
1    abc    and    None  1992     F
     efg    and    Blue  2000  None
     fgs    and     Red  2001  None
2    dcv    and    None  2001     F
     hij    and    None  2002  None
     tyu    and  Yellow  1999  None
3    jhg    and    None  2001     M
4    mlh    and    None  2001  None

It is a good practice when you create a new index to always check it is unique:

df.index.is_unique
True

Data from file

If your data are in a file, you should first import it into a variable using PSL json module:

import json
with open(path) as file:
    data = json.load(file)

This will do the trick, then come back to the beginning of my answer.

Answered By - jlandercy