CODEWITHSUNDEEP
pythonpandasdataframetype-conversionpivot-table
jurkij
jurkij

Reputation: 155

Pandas Pivot table converts float to int

I found weird behavior of pandas when converting data frame to pivot table.

import pandas as pd
df = pd.DataFrame({'car_id': {0: 'Trabant', 1: 'Buick', 2: 'Dodge'}, 'car_order': {0: 2, 1: 1, 2: 14}, 'car_name': {0: 'Trabant', 1: 'Buick', 2: 'Dodge'}, 'car_rank': {0: 111111317.29, 1: 1111112324.0, 2: 1111112324.5}})
table = df.pivot_table(index=['car_id', 'car_name', 'car_order'], columns=[],values=['car_rank'], fill_value='',dropna=True)
print table

df1 = pd.DataFrame({'car_id': {0: 'Trabant', 1: 'Buick', 2: 'Dodge'}, 'car_order': {0: 2, 1: 1, 2: 14}, 'car_name': {0: 'Trabant', 1: 'Buick', 2: 'Dodge'}, 'car_rank': {0: 17.29, 1: 24.0, 2: 24.5}})
table1 = df1.pivot_table(index=['car_id', 'car_name', 'car_order'], columns=[],values=['car_rank'], fill_value='',dropna=True)
print table1

Result output:

Table
                              car_rank
car_id  car_name car_order            
Buick   Buick    1          1111112324
Dodge   Dodge    14         1111112324
Trabant Trabant  2           111111317

Table 1
                            car_rank
car_id  car_name car_order          
Buick   Buick    1             24.00
Dodge   Dodge    14            24.50
Trabant Trabant  2             17.29

Do you know why in Table are values converted to int and for Table 1 values stay as float?

pandas 0.18.0 , python 2.7.9

Upvotes: 5

Views: 7657

Answers (1)

MaxU - stand with Ukraine
MaxU - stand with Ukraine

Reputation: 210852

here is result of my observations of pandas 0.18.0:

Source code of pandas/tools/pivot.py definition of pivot_table() lines: 141-142:

if fill_value is not None:
    table = table.fillna(value=fill_value, downcast='infer')

This is exactly what happened to your pivoted DF: 

In [78]: df.fillna('', downcast='infer')
Out[78]:
    car_id car_name  car_order    car_rank
0  Trabant  Trabant          2   111111317
1    Buick    Buick          1  1111112324
2    Dodge    Dodge         14  1111112324

Types:

In [48]: df.fillna('', downcast='infer').dtypes
Out[48]:
car_id       object
car_name     object
car_order     int64
car_rank      int64
dtype: object

Interestingly enough - if you use pivot_table() properly (i.e. for pivoting) - it works correctly:

In [81]: df.pivot_table(index=['car_id', 'car_order'], columns=['car_name'], values=['car_rank'],dropna=True, fill_value='')
Out[81]:
                       car_rank
car_name                  Buick         Dodge      Trabant
car_id  car_order
Buick   1         1111112324.00
Dodge   14                      1111112324.50
Trabant 2                                     111111317.29

PS I still can't understand why are you using pivot_table in that strange way - what are you going to achieve?

Upvotes: 6

Related Questions