8000 ENH: Add Series.set_index #27504
ENH: Add Series.set_index #27504
Conversation
|
I think this has the same issue as the previous PR where the semantics for a Series now break from a DataFrame as the former here accepts list-likes whereas the latter is used for labels. Haven't kept totally current on this - was this approach ever agreed to in one of the issues or are you just submitting to restart conversation? |
|
So to maybe get you up to speed here the semantics of this PR would differ from a DataFrame. Taking your first test case this: # Sets the index as "A", "B", "C"
>>> pd.Series(range(3)).set_index(["A", "B", "C"])
A 0
B 1
C 2
dtype: int64Would exhibit different behavior than what happens on a DataFrame with the same number of rows # Selects columns "A", "B", "C" and moves to the index
>>> pd.DataFrame(np.ones((3,3)), columns=list('ABC')).set_index(["A", "B", "C"])
Empty DataFrame
Columns: []
Index: [(1.0, 1.0, 1.0), (1.0, 1.0, 1.0), (1.0, 1.0, 1.0)]So in a vacuum this PR would make sense but given the current state of things would need to have something additional to bridge behavior between the two object types. |
|
This is very contentious issue among the maintainers it seems, and considering how #22225 ended, seems unlikely this is going anywhere. |
DataFrame.set_indexhas)Help users method-chain all the things. This is modeled closely on the original support for Arrays added in 2011 by Wes Mckinney in
553bacc
ENH: can pass external arrays in lieu of column names to DataFrame.set_indexThis is also how
DataFrame.set_indexbehaves (though poorly documented), modulo support for column keys which don't apply here.related discussion #24046, and end of #14829 (comment), #12392