python - Pivoting a Pandas Dataframe containing strings - 'No numeric types to aggregate' error

Question

There is a good number of questions about this error, but after looking around I'm still not able to find/wrap my mind around a solution yet. I'm trying to pivot a data frame with strings, to get some row data to become columns, but not working out so far.

Shape of my df

<class 'pandas.core.frame.DataFrame'>
Int64Index: 515932 entries, 0 to 515931
Data columns (total 5 columns):
id                 515932 non-null object
cc_contact_id      515932 non-null object
Network_Name       515932 non-null object
question           515932 non-null object
response_answer    515932 non-null object
dtypes: object(5)
memory usage: 23.6+ MB

Sample format

id  contact_id  question    response_answer
16  137519  2206    State   Ca
17  137520  2206    State   Ca
18  137521  2206    State   Ca
19  137522  2206    State   Ca
20  137523  2208    City    Lancaster
21  137524  2208    City    Lancaster
22  137525  2208    City    Lancaster
23  137526  2208    City    Lancaster
24  137527  2208    Trip_End Location   Home
25  137528  2208    Trip_End Location   Home
26  137529  2208    Trip_End Location   Home
27  137530  2208    Trip_End Location   Home

What I would like to pivot to

id  contact_id      State   City       Trip_End Location
16  137519  2206    Ca      None       None None
20  137523  2208    None    Lancaster  None None
24  137527  2208    None    None       None Home
etc. etc. 

Where the question values become the columns, with the response_answer being in it's corresponding column, and retaining the ids

What I have tried

unified_df = pd.DataFrame(unified_data, columns=target_table_headers, dtype=object)

pivot_table = unified_df.pivot_table('response_answer',['id','cc_contact_id'],'question')
# OR
pivot_table = unified_df.pivot_table('response_answer','question')

DataError: No numeric types to aggregate

What is the way to pivot a data frame with string values?

Answer 1

The default aggfunc in pivot_table is np.sum and it doesn't know what to do with strings and you haven't indicated what the index should be properly. Trying something like:

pivot_table = unified_df.pivot_table(index=['id', 'contact_id'],
                                     columns='question', 
                                     values='response_answer',
                                     aggfunc=lambda x: ' '.join(x))

This explicitly sets one row per id, contact_id pair and pivots the set of response_answer values on question. The aggfunc just assures that if you have multiple answers to the same question in the raw data that we just concatenate them together with spaces. The syntax of pivot_table might vary depending on your pandas version.

Here's a quick example:

In [24]: import pandas as pd

In [25]: import random

In [26]: df = pd.DataFrame({'id':[100*random.randint(10, 50) for _ in range(100)], 'question': [str(random.randint(0,3)) for _ in range(100)], 'response': [str(random.randint(100,120)) for _ in range(100)]})

In [27]: df.head()
Out[27]:
     id question response
0  3100        1      116
1  4500        2      113
2  5000        1      120
3  3900        2      103
4  4300        0      117

In [28]: df.info()
<class 'pandas.core.frame.DataFrame'>
Int64Index: 100 entries, 0 to 99
Data columns (total 3 columns):
id          100 non-null int64
question    100 non-null object
response    100 non-null object
dtypes: int64(1), object(2)
memory usage: 3.1+ KB

In [29]: df.pivot_table(index='id', columns='question', values='response', aggfunc=lambda x: ' '.join(x)).head()
Out[29]:
question        0        1    2        3
id
1000      110 120      NaN  100      NaN
1100          NaN  106 108  104      NaN
1200      104 113      119  NaN      101
1300          102      NaN  116  108 120
1400          NaN      NaN  116      NaN