pandas.core.groupby.DataFrameGroupBy to_csv method doesn't ouput csv file as expected

[c0indev3l]

>>> df1 = pd.DataFrame( { 
    "Name" : ["Alice", "Bob", "Mallory", "Mallory", "Bob" , "Mallory"] , 
    "City" : ["Seattle", "Seattle", "Portland", "Seattle", "Seattle", "Portland"] } )

>>> g1 = df1.groupby( [ "Name" ] )

>>> print g1.head()
               City     Name
Name                        
Alice   0   Seattle    Alice
Bob     1   Seattle      Bob
        4   Seattle      Bob
Mallory 2  Portland  Mallory
        3   Seattle  Mallory
        5  Portland  Mallory

>>> g1.to_csv('out.csv')
g1.to_csv('out.csv')
Out[10]: 
Name
Alice      None
Bob        None
Mallory    None
dtype: object

(Why some data are output to ipython console ?)

>>> !cat out.csv
,City,Name
2,Portland,Mallory
3,Seattle,Mallory
5,Portland,Mallory

[cpcloud]

What is your end goal here? You shouldn't really be using to_csv on a groupby. If you really want to write each group separately or you need to do some processing on each group before writing, consider looping over the groups:

for group_name, df in df.groupby('Name'):
    newdf = process(df)
    with open('the_csv.csv', 'a') as f:
        df.to_csv(f)

[jreback]

If you REALLY want the output you have, you can do this, but as @cpcloud , I don't see utility in this

In [46]: df1.reset_index().set_index(['Name','City']).sortlevel(0)
Out[46]: 
                  index
Name    City           
Alice   Seattle       0
Bob     Seattle       1
        Seattle       4
Mallory Portland      2
        Portland      5
        Seattle       3

In [47]: df1.reset_index().set_index(['Name','City']).sortlevel(0).to_csv('test.csv')

In [48]: !cat test.csv
Name,City,index
Alice,Seattle,0
Bob,Seattle,1
Bob,Seattle,4
Mallory,Portland,2
Mallory,Portland,5
Mallory,Seattle,3

[c0indev3l]

This issue is linked to #4883

[jtratner]

Why does a group by object even have a to_csv method?

[jreback]

it doesn't, its dispatching to the object (which is how apply works)

[cpcloud]

Because it creates a wrapper for the type of groupby lazily

[jreback]

I suppose should explicity allow certain methods on groupby

[cpcloud]

maybe although special casing everything could turn into a mess

[cpcloud]

well not everything but none of the IO stuff really makes sense on a groupby

[jtratner]

Yeah not worth the time.

[jtratner]

One thing we could do is implement dir on the object so it's clearer
what's available for tab completion and introspection.

[cpcloud]

let me take a look at how that wrapper is constructed....it might be deep in some lambda somewhere

[jreback]

closed by #4887