To facilitate groupby-type operations, from mailing list:

With the following:

df # predefined
df['E'] = pandas.Series(0.)
for grp_name, grp_data in df.groupby(['C','D']):
   # snipped code here; returns a scalar x that is dependent on
grp_data
   grp_data['E'] = x   # NB df['E'] not updated

I was wondering if there was a way of internally capturing child-
parent relationships between dataframes so that each 'grp_data'
dataframe knows that its parent is dataframe 'df' and knows how to
propagate assignments back to the parent dataframe using column names
and row masks.  I suppose this would require that grp_data is an
instance of a subclass of pandas.DataFrame.```

Not sure how feasible this is, but an idea. 

Comment From: wesm

Closing in favor of more generalized work on deferred evaluation in pandas 2.0