Pandas: How to select the entry with the lowest mean per group?

Suppose I have the following dataframe

df = pd.DataFrame([
    (2, 2, 'A', .5),
    (2, 2, 'A', .6),
    (2, 2, 'B', .75),
    (2, 2, 'B', .7),
    (2, 2, 'C', .6),

    (2, 3, 'A', .65),
    (2, 3, 'A', .6),
    (2, 3, 'B', .75),
    (2, 3, 'B', .7),
    (2, 3, 'C', .6)
], columns=['out_size', 'problem_size', 'algo', 'time'])

I want to

• group by `[out_size’, ‘problem_size’, ‘algo’], and for each group
• count the number of occurrences for each algo, then
• select/keep the algo that has the lowest average time in that group,

result

pd.DataFrame(
      [[2, 2, 'A', 0.55],
       [2, 3, 'C', 0.6]], columns=['out_size', 'problem_size', 'algo', 'time'])

>Solution :

You can use a double groupby:

cols = ['out_size', 'problem_size', 'algo']

out = (df
 .groupby(cols, as_index=False)['time'].mean()
 .sort_values(by='time')
 .groupby(cols[:-1], as_index=False).first()
)

Slightly more efficient alternative that doesn’t require to sort the values (but requires to store an intermediate):

cols = ['out_size', 'problem_size', 'algo']

out = df.groupby(cols)['time'].mean()
out = out.loc[out.groupby(cols[:-1]).idxmin()].reset_index()

output:

   out_size  problem_size algo  time
0         2             2    A  0.55
1         2             3    C  0.60