Pandas: parse merged header columns from Excel

Samarth Bharadwaj picture Samarth Bharadwaj · Dec 11, 2014 · Viewed 9.6k times · Source

The data in excel sheets is stored as follows:

   Area     |          Product1     |      Product2        |      Product3
            |      sales|sales.Value|   sales |sales.Value |  sales |sales.Value
  Location1 |    20     | 20000     |      25 |  10000     |   200  | 100
  Location2 |    30     | 30000     |      3  | 12300      |   213  | 10

the product name is a merge of 2 cells of two rows "no of sales" and "sales value" for each of 1000 or so areas for a given month. Similarly there are separate files for each month for the last 5 years. Further, new products have been added and removed in different months. So a different month file might look like:

   Area     |          Product1     |      Product4        |      Product3

Can the forum suggest the best way to read this data using pandas? I can't use index since the product columns are different each month

Ideally, I would like to convert the initial format above to:

 Area      | Product1.sales|Product1.sales.Value| Product2.sales |Product2.sales.Value | 
 Location1 | 20            | 20000              | 25             | 10000               |  
 Location2 | 30            | 30000              | 3              | 12300               | 

import pandas as pd
xl_file = read_excel("file path", skiprow=2, sheetname=0)
/* since the first two rows are always blank */


                  0            1        2               3                      4
      0          NaN          NaN      NaN       Auto loan                    NaN
      1  Branch Code  Branch Name   Region  No of accounts  Portfolio Outstanding
      2         3000       Name1  Central               0                      0
      3         3001       Name2  Central               0                      0

I want to convert it to Auto loan.No of account, Auto loan.Portfolio Outstanding as the headers.

Answer

unutbu picture unutbu · Dec 11, 2014

Suppose your DataFrame is df:

import numpy as np
import pandas as pd

nan = np.nan
df = pd.DataFrame([
    (nan, nan, nan, 'Auto loan', nan)
    , ('Branch Code', 'Branch Name', 'Region', 'No of accounts'
       , 'Portfolio Outstanding')
    , (3000, 'Name1', 'Central', 0, 0)
    , (3001, 'Name2', 'Central', 0, 0)
])

so that it looks like this:

             0            1        2               3                      4
0          NaN          NaN      NaN       Auto loan                    NaN
1  Branch Code  Branch Name   Region  No of accounts  Portfolio Outstanding
2         3000       Name1  Central               0                      0
3         3001       Name2  Central               0                      0

Then first forward fill the NaNs in the first two rows (thus propagating 'Auto loan', for example).

df.iloc[0:2] = df.iloc[0:2].fillna(method='ffill', axis=1)

Next fill in the remaining NaNs with empty strings:

df.iloc[0:2] = df.iloc[0:2].fillna('')

Now join the two rows together with . and assign that as the column level values:

df.columns = df.iloc[0:2].apply(lambda x: '.'.join([y for y in x if y]), axis=0)

And finally, remove the first two rows:

df = df.iloc[2:]

This yields

  Branch Code Branch Name   Region Auto loan.No of accounts  \
2        3000      Name1  Central                        0   
3        3001      Name2  Central                        0   

  Auto loan.Portfolio Outstanding  
2                               0  
3                               0  

Alternatively, you could create a MultiIndex column instead of creating a flat column index:

import numpy as np
import pandas as pd

nan = np.nan
df = pd.DataFrame([
    (nan, nan, nan, 'Auto loan', nan)
    , ('Branch Code', 'Branch Name', 'Region', 'No of accounts'
       , 'Portfolio Outstanding')
    , (3000, 'Name1', 'Central', 0, 0)
    , (3001, 'Name2', 'Central', 0, 0)
])
df.iloc[0:2] = df.iloc[0:2].fillna(method='ffill', axis=1)
df.iloc[0:2] = df.iloc[0:2].fillna('Area')

df.columns = pd.MultiIndex.from_tuples(
    zip(*df.iloc[0:2].to_records(index=False).tolist()))
df = df.iloc[2:]

Now df looks like this:

         Area                           Auto loan                      
  Branch Code Branch Name   Region No of accounts Portfolio Outstanding
2        3000      Name1  Central              0                     0
3        3001      Name2  Central              0                     0

the column is a MultiIndex:

In [275]: df.columns
Out[275]: 
MultiIndex(levels=[[u'Area', u'Auto loan'], [u'Branch Code', u'Branch Name', u'No of accounts', u'Portfolio Outstanding', u'Region']],
           labels=[[0, 0, 0, 1, 1], [0, 1, 4, 2, 3]])

The column has two levels. The first level has values [u'Area', u'Auto loan'], the second has values [u'Branch Code', u'Branch Name', u'No of accounts', u'Portfolio Outstanding', u'Region'].

You can then access a column by specifing the value from both levels:

print(df.loc[:, ('Area', 'Branch Name')])
# 2    Name1
# 3    Name2
# Name: (Area, Branch Name), dtype: object

print(df.loc[:, ('Auto loan', 'No of accounts')])
# 2    0
# 3    0
# Name: (Auto loan, No of accounts), dtype: object

One advantage of using a MultiIndex is that you can easily select all columns which have a certain level value. For instance, to select the sub-DataFrame having to do with Auto loans you could use:

In [279]: df.loc[:, 'Auto loan']
Out[279]: 
  No of accounts Portfolio Outstanding
2              0                     0
3              0                     0

For more on selecting rows and columns from a MultiIndex, see MultiIndexing Using Slicers.