CODEWITHSUNDEEP
pythonpandasdataframepandas-groupby
jbeckom
jbeckom

Reputation: 105

Pandas DataFrame GroupBy Rank

DataFrame:

        account_id  plan_id     policy_group_nbr    plan_type               split_eff_date  splits
0       470804      8739131     Conversion732       Onsite Medical Center   1/19/2022       Bob Smith (28.2) | John Doe (35.9) | A...
1       470804      8739131     Conversion732       Onsite Medical Center   1/21/2022       Bob Smith (19.2) | John Doe (34.6) | A...
2       470809      2644045     801790              401(k)                  1/18/2022       Jim Jones (100)
3       470809      2644045     801790              401(k)                  1/5/2022        Other Name (50) | Jim Jones (50)
4       470809      2738854     801789              401(k)                  1/18/2022       Jim Jones (100)
... ... ... ... ... ... ...
1720    3848482     18026734    24794               Accident                1/20/2022       Bill Underwood (50) | Jim Jones (50)
1721    3848482     18026781    BCSC                FSA Admin               1/20/2022       Bill Underwood (50) | Jim Jones (50)
1722    3927880     19602958    Consulting          Other                   1/20/2022       Bill Brown (50) | Tim Scott (50)
1723    3927880     19863300    Producer Expense    5500 Filing             1/20/2022       Bill Brown (50) | Tim Scott (50)
1724    3927880     19863300    Producer Expense    5500 Filing             1/21/2022       Bill Brown (50) | Tim Scott (50)

I need to group by (account_id, plan_id, policy_group_nbr, plan_type) sorted by split_eff_date (desc), in order to remove all rows for the group but the most recent date while maintaining all columns. I can get a rank however, when attempting to pass an argument to the lambda function, I'm receiving a TypeError.

working as expected:

splits['rank'] = splits.groupby(['account_id', 'plan_id', 'policy_group_nbr', 'plan_type'])['split_eff_date'].apply(lambda x: x.sort_values().rank())

TypeError: incompatible index of inserted column with frame index

splits['rank'] = splits.groupby(['account_id', 'plan_id', 'policy_group_nbr', 'plan_type'])['split_eff_date'].apply(lambda x: x.sort_values(ascending=False).rank())

passing the axis argument didn't seem to help either... is this a simple syntax issue, or am I not understanding the function properly?

Upvotes: 2

Views: 477

Answers (1)

mechanical_meat
mechanical_meat

Reputation: 169284

easier -- and typically faster -- to do this with .transform().

easier because when you sort descending, the index doesn't match when you try to assign back to original DataFrame. i tried not using an index in the .groupby(), but wasn't able to get that working.

link to documentation about .transform(): https://pandas.pydata.org/docs/reference/api/pandas.core.groupby.SeriesGroupBy.transform.html

i recommend using .transform() like this, and be sure to supply ascending=False kwarg to .rank() as well:

df["rank2"] = df.groupby(["account_id", "plan_id", "policy_group_nbr", "plan_type"])[ 
    "split_eff_date" 
].transform( 
    lambda x: x.sort_values(ascending=False).rank(ascending=False, method="first") 
)

result with both kinds of ranking -- i took just the first 5 rows from your sample data:

In [93]: df
Out[93]: 
   account_id  plan_id policy_group_nbr              plan_type split_eff_date  rank  rank2
3      470809  2644045           801790                 401(k)     2022-01-05   1.0    2.0
2      470809  2644045           801790                 401(k)     2022-01-18   2.0    1.0
4      470809  2738854           801789                 401(k)     2022-01-18   1.0    1.0
0      470804  8739131    Conversion732  Onsite Medical Center     2022-01-19   1.0    2.0
1      470804  8739131    Conversion732  Onsite Medical Center     2022-01-21   2.0    1.0

Upvotes: 1

Related Questions