How to extract the n-th maximum/minimum value in a column of a DataFrame in pandas?

Question

I would like to obtain the n-th minimum or the n-th maximum value from numerical columns in the DataFrame in pandas.

Example:

df = pd.DataFrame({'a': [3.0, 2.0, 4.0, 1.0],'b': [1.0, 4.0 , 2.0, 3.0]})

     a    b
0  3.0  1.0
1  2.0  4.0
2  4.0  2.0
3  1.0  3.0

The third largest value in column a is 2 and the second smallest value in column b is also 2.

cs95 · Accepted Answer · 2017-12-29T18:09:05.137

29

You can use nlargest/nsmallest -

df    
     a    b
0  3.0  1.0
1  2.0  4.0
2  4.0  2.0
3  1.0  3.0

df.a.nlargest(3).iloc[-1]
2.0

Or,

df.a.nlargest(3).iloc[[-1]]

1    2.0
Name: a, dtype: float64

And, as for b -

df.b.nsmallest(2).iloc[-1]
2.0

Or,

df.b.nsmallest(2).iloc[[-1]]

2    2.0
Name: b, dtype: float64

Quick observation here - this sort of operation cannot be vectorised. You are essentially performing two completely different operations here.

edited Dec 29 '17 at 18:09

answered Dec 29 '17 at 17:55

cs95

@KrzysztofSłowiński You're welcome. Edited my answer. – cs95 Dec 29 '17 at 18:09
1

@COLDSPEED it there a way to return only a single n-th maximum / minimum row or the approach is to use nlargest / nsmallest and then take the last row? – Krzysztof Słowiński Dec 30 '17 at 19:58
@KrzysztofSłowiński the other alternative is calling x.sort_values() and accessing the Nth largest/smallest value depending on whether you sort in ascending or descending. Either way, you must sort. – cs95 Dec 30 '17 at 20:02
Sure, I was thinking if there was a function or a recommended way. – Krzysztof Słowiński Dec 30 '17 at 20:04
1

@KrzysztofSłowiński as of now, there is only nlargest, but there could be a nThLargest in the future :-) – cs95 Dec 30 '17 at 20:07
Yes, exactly, if it is a common use case. – Krzysztof Słowiński Dec 30 '17 at 20:08

score 5 · Answer 2 · edited Jun 04 '18 at 07:16

5

df =  
     a    b
0  3.0  1.0
1  2.0  4.0
2  4.0  2.0
3  1.0  3.0

df.nlargest(3,'a')
   =2.0

df.nsmallest(2,'b')=2.0

edited Jun 04 '18 at 07:16

cs95

answered Dec 30 '17 at 00:39

nbm

score 1 · Answer 3 · answered Jun 30 '22 at 09:00

1

5th max: df.a.nlargest(5).min()
5th min: df.a.nsmallest(5).max().
Tips: add param keep may acts better:
df.a.nlargest(5,keep='first').min()

answered Jun 30 '22 at 09:00

Anthony

This question is relatively old so thank you for this solution because it seems to be the most pythonic and clean (compared to others). How about performance? What is your experience? – Soren V. Raben Jul 23 '23 at 13:11

3 Answers3