How to implement multiple aggregations using pandas groupby, referencing a specific column - pandas

Question

I have data in a pandas data frame, and need to aggregate it. I need to do different aggregations across different columns similar to the below.
group min(rank) min(rank) min sum
title t_no t_descr rank stores
A 1 a 1 1000
B 1 a 1 1000
B 2 b 2 800
C 2 b 2 800
D 1 a 1 1000
D 2 b 2 800
F 4 d 4 500
E 3 c 3 700
to:
title t_no t_descr rank stores
A 1 a 1 1000
B 1 a 1 1800
C 2 b 2 800
D 1 a 1 1800
E 3 c 3 700
F 4 d 4 500
You'll notice that title B and D have been aggregated, keeping the t_no & t_descr that corresponded to the minimum of the rank for the respective title group, while stores are summed. t_no & t_descr are just arbitrary text. I need the top rank by title, sum the stores, and keep the corresponding t_no & t_descr.
How can I do this within a single pandas groupby? This is dummy data; the real problem that I'm working on has many more aggregations, and I'd prefer not to have to do each aggregation individually, which I know how to do.
I started with the below, but realized that I really need the mins & maxs for t_no & t_descr to be based on rank col of the subgroup, not the columns themselves.
aggs = {
'rank': 'min',
't_no': 'min', # need t_no for row that is min(rank) by title.
't_descr': 'min' # need t_descr for row that is min(rank) by title.
}
df2.groupby('title').agg(aggs).reset_index()
Perhaps there's a way to do this with a lambda? I'm sure there's a straightforward way to do this. And if groupby isn't the right method I'm obviously open to suggestions.
Thanks!

Answer 1

Two step process...
aggregate for sum of stores and idxmin for rank...
then use idxmin to slice original dataframe and join it with the aggregate
agged = df.groupby('title').agg(dict(rank='idxmin', stores='sum'))
df.loc[agged['rank'], ['title', 't_no', 't_descr', 'rank']].join(agged.stores, on='title')
title t_no t_descr rank stores
0 A 1 a 1 1000
1 B 1 a 1 1800
3 C 2 b 2 800
4 D 1 a 1 1800
7 E 3 c 3 700
6 F 4 d 4 500

Answer 2

This is a slightly different approach from #piRSquared, but gets you to the same spot:
Code:
# Set min and sum functions according to columns and generate new dataframe
f = {'rank':min, 'rank':min, 'stores':sum}
grouped = df.groupby('title').agg(f).reset_index()
# Then merge with original dataframe (keeping only the merged and new columns)
pd.merge(grouped, df[['title','rank','t_no','t_descr']], on=['title','rank'])
Output:
title stores rank t_no t_descr
0 A 1000 1 1 a
1 B 1800 1 1 a
2 C 800 2 2 b
3 D 1800 1 1 a
4 E 700 3 3 c
5 F 500 4 4 d
Of course you can organize the columns as you see fit.