pivot_table returns a Series instead of a DataFrame depending upon the datatype of the values parameter

[vijaymukhi712]

My python code looks like

df = pandas.read_csv("a.csv")
print df
table = pandas.pivot_table(df, values=['Sales1'] ,rows='State' , aggfunc=np.sum)
print type(table)
table = pandas.pivot_table(df, values='Sales1' ,rows='State' , aggfunc=np.sum)
print type(table)

Contents of a.csv
City,Sales1
Mumbai,1

The Output
City  Sales1
0  Mumbai       1
<class 'pandas.core.frame.DataFrame'>
<class 'pandas.core.series.Series'>

My question is why would a change in the datatype of the values parameter return a DataFrame instead of a Series. After all we use an array only if we have to pass multiple values.

Thank You.

Vijay Mukhi

[cpcloud]

Hi @vijaymukhi712,

Thanks for the report. Can you please post a minimal reproducible example, meaning that anyone who wants to could load your code into Python and read in a data set that illustrates the issue? It doesn't have to be the data set you found the issue with, but that's okay if you want too.

The issue will get resolved much faster if you do this.

Thanks.

[vijaymukhi712]

I am sorry for not being clear earlier. I have created the smallest csv, a.csv file that shows the anomaly. The data set that I originally worked with was over 80 MB, so no point posting it. As the example above shows why would changing the data type of the values parameter from a string to an array change the data type of the returned object from a Series to a DataFrame. That is the basic question I am asking. My example works with any data set that has two columns and a minimum of one row.

[TomAugspurger]

@vijaymukhi712 , when @cpcloud asked for a minimal reproducible example, something like df = pd.DataFrame({'A': [1, 2, 3, 4], 'State': ['IA', 'NY', 'NY', 'CA']}) would be fantastic. Just something that can be copy-pasted directly into the interpreter.

I think that values=['Sales1'] returning a DataFrame while values='Sales1' returns a Series is the intended behavior. ['Sales1'] is passing a collection of objects, which must be represented by a DataFrame. It just happens that the DataFrame only has one column since the collection only had one item. You'll notice the same thing when selecting subsets of columns of a DataFrame.

In [13]: type(df['State'])
Out[13]: pandas.core.series.Series

In [14]: type(df[['State']])
Out[14]: pandas.core.frame.DataFrame

Here the outer square brackets is the getitem, and in the second case the inner square brackets create a list with one element.

So you're assumption here

After all we use an array only if we have to pass multiple values.

was faulty. Sometimes we do pass arrays with single values.

[cpcloud]

@TomAugspurger nice explanation

[vijaymukhi712]

Tom, thanks for the insight, never looked at that way.