CODEWITHSUNDEEP
pythonpandaspandas-groupbyseriesmulti-index
Brandon Loyal Sorensen
Brandon Loyal Sorensen

Reputation: 53

Getting the nlargest of each group in a Multiindex Pandas Series

I have a DataFrame that consists of information about every NFL play that has occurred since 2009. My goal is to find out which teams had the most "big plays" in each season. To do this, I found all plays which gained over 20 yards, grouped them by year and team, and got the size of each of those group.

big_plays = (df[df['yards_gained'] >= 20]
             .groupby([df['game_date'].dt.year, 'posteam'])
             .size())

This results in the following Series:

game_date  posteam
2009       ARI        55
           ATL        51
           BAL        55
           BUF        37
           CAR        52
           CHI        58
           CIN        51
           CLE        31
           DAL        68
           DEN        42
           DET        42
           GB         65
           HOU        63
           IND        67
           JAC        51
           KC         44
           MIA        34
           MIN        64
           NE         48
           NO         72
           NYG        69
           NYJ        54
           OAK        38
           PHI        68
           PIT        72
           SD         71
           SEA        45
           SF         51
           STL        42
           TB         51
                      ..
2018       BAL        44
           BUF        55
           CAR        64
           CHI        66
           CIN        69
           CLE        70
           DAL        51
           DEN        59
           DET        51
           GB         63
           HOU        53
           IND        57
           JAX        51
           KC         88
           LA         80
           LAC        77
           MIA        47
           MIN        56
           NE         64
           NO         66
           NYG        70
           NYJ        49
           OAK        63
           PHI        54
           PIT        66
           SEA        62
           SF         69
           TB         73
           TEN        51
           WAS        46
Length: 323, dtype: int64

So far, this is exactly what I want. However, I am stuck on the next step. I want the n-largest values for each group in the MultiIndex, or the n-teams with the most number of "big plays" per season.

I have semi-successfully solved this task in a cumbersome way. If I groupby the 0th level of the MultiIndex, then run the nlargest function on that groupby, I get the following (truncated to the first two years for brevity):

big_plays.groupby(level=0).nlargest(5)

returns

game_date  game_date  posteam
2009       2009       NO         72
                      PIT        72
                      SD         71
                      NYG        69
                      DAL        68
2010       2010       PHI        81
                      NYG        78
                      PIT        78
                      SD         75
                      DEN        73

This (rather inelegantly) solves the problem, but I'm wondering how I can better achieve more or less the same results.

Upvotes: 5

Views: 1161

Answers (1)

jezrael
jezrael

Reputation: 862691

In my opinion your code is nice, only a bit changed by group_keys=False in Series.groupby for avoid duplicated MultiIndex levels:

s = big_plays.groupby(level=0, group_keys=False).nlargest(5)
print (s)
game_date  posteam
2009       NO         72
           PIT        72
           SD         71
           NYG        69
           DAL        68
2018       KC         88
           LA         80
           LAC        77
           TB         73
           CLE        70
Name: a, dtype: int64

df = big_plays.groupby(level=0, group_keys=False).nlargest(5).reset_index(name='count')
print (df)
   game_date posteam  count
0       2009      NO     72
1       2009     PIT     72
2       2009      SD     71
3       2009     NYG     69
4       2009     DAL     68
5       2018      KC     88
6       2018      LA     80
7       2018     LAC     77
8       2018      TB     73
9       2018     CLE     70

Alternative is more complicated:

df = (big_plays.reset_index(name='count')
               .sort_values(['game_date','count'], ascending=[True, False])
               .groupby('game_date')
               .head(5))
print (df)
    game_date posteam  count
19       2009      NO     72
24       2009     PIT     72
25       2009      SD     71
20       2009     NYG     69
8        2009     DAL     68
43       2018      KC     88
44       2018      LA     80
45       2018     LAC     77
57       2018      TB     73
35       2018     CLE     70

Upvotes: 4

Related Questions