python - select largest N of a column of each groupby group using pandas

Question

My df:

{'city1': {0: 'Chicago',
  1: 'Chicago',
  2: 'Chicago',
  3: 'Chicago',
  4: 'Miami',
  5: 'Houston',
  6: 'Austin'},
 'city2': {0: 'Toronto',
  1: 'Detroit',
  2: 'St.Louis',
  3: 'Miami',
  4: 'Dallas',
  5: 'Dallas',
  6: 'Dallas'},
 'p234_r_c': {0: 5.0, 1: 4.0, 2: 2.0, 3: 0.5, 4: 1.0, 5: 4.0, 6: 3.0},
 'plant1_type': {0: 'COMBCYCL',
  1: 'COMBCYCL',
  2: 'NUKE',
  3: 'COAL',
  4: 'NUKE',
  5: 'COMBCYCL',
  6: 'COAL'},
 'plant2_type': {0: 'COAL',
  1: 'COAL',
  2: 'COMBCYCL',
  3: 'COMBCYCL',
  4: 'COAL',
  5: 'NUKE',
  6: 'NUKE'}}

I want to do 2 groupby operations and take the largest 1 of each group using column p234_r_c.

1st groupby = ['plant1_type', 'plant2_type', 'city1']

2nd groupby = ['plant1_type', 'plant2_type', 'city2']

As such I do the following:

df.groupby(['plant1_type','plant2_type','city1'])['p234_r_c'].
    nlargest(1).reset_index()


plant1_type plant2_type city1   level_3 p234_r_c
0   COAL    COMBCYCL    Chicago 3   0.5
1   COAL    NUKE        Austin  6   3.0
2   COMBCYCL    COAL    Chicago 0   5.0
3   COMBCYCL    NUKE    Houston 5   4.0
4   NUKE    COAL        Miami   4   1.0
5   NUKE    COMBCYCL    Chicago 2   2.0

The result of the 1st groupby makes sense. However, I am confused by the result of the 2nd groupby:

df.groupby(['plant1_type','plant2_type','city2'])['p234_r_c'].
    nlargest(1).reset_index()

index   p234_r_c
0   0   5.0
1   1   4.0
2   2   2.0
3   3   0.5
4   4   1.0
5   5   4.0
6   6   3.0

What happened to columns plant1_type, plant2_type and city2 in the result? Shouldnt they appear in the result just like how plant1_type, plant2_type and city1 appeared in the result of the 1st groupby?

Answer 1

I added an issue here

Theory:

When the results of a groupby on a pd.Series returns the same pd.Series values, then the original index is returned.

Boiled down example

df = pd.DataFrame(dict(A=[0, 1, 2, 3]))

# returns results identical to df.A
print(df.groupby(df.A // 2).A.nsmallest(2))

# returns results out of order
print(df.groupby(df.A // 2).A.nlargest(2))

0    0
1    1
2    2
3    3
Name: A, dtype: int64
A   
0  1    1
   0    0
1  3    3
   2    2
Name: A, dtype: int64

I'd argue that you want these to return the same consistent index.

This is the most egregious consequence of this:

# most egregious
# this will be randomly different
print(df.groupby(df.A // 2).A.apply(pd.Series.sample, n=2))

---

returns this on one execution

A   
0  1    1
   0    0
1  2    2
   3    3
Name: A, dtype: int64

And this on another

0    0
1    1
2    2
3    3
Name: A, dtype: int64

Of course this never has an issue because it's impossible to return the same values as the original

print(df.groupby(df.A // 2).A.apply(pd.Series.sample, n=1))

A   
0  0    0
1  2    2
Name: A, dtype: int64

---

Work around
set_index

cols = ['plant1_type','plant2_type','city2']
df.set_index(cols).groupby(level=cols)['p234_r_c'].
    nlargest(1).reset_index()

  plant1_type plant2_type     city2  p234_r_c
0    COMBCYCL        COAL   Toronto       5.0
1    COMBCYCL        COAL   Detroit       4.0
2        NUKE    COMBCYCL  St.Louis       2.0
3        COAL    COMBCYCL     Miami       0.5
4        NUKE        COAL    Dallas       1.0
5    COMBCYCL        NUKE    Dallas       4.0
6        COAL        NUKE    Dallas       3.0