Sorting
df.sort_values('revenue', ascending=False)
df.sort_values(['region', 'revenue'], ascending=[True, False])
df.sort_values('revenue', na_position='first')Remember to assign the result — df.sort_values(...) alone changes nothing.
Top N
df.nlargest(10, 'revenue')
df.nsmallest(5, 'revenue')Faster and clearer than sorting the whole frame and slicing, especially on large data.
Ranking
df['rank'] = df['revenue'].rank(ascending=False)
df['rank'] = df['revenue'].rank(method='dense', ascending=False)| method | Behaviour on ties |
|---|---|
average | Default — ties share the mean rank, e.g. 2.5 |
min | Ties get the lowest rank, then skip — like SQL RANK |
dense | Ties share, no skip — like SQL DENSE_RANK |
first | Broken by order of appearance — like ROW_NUMBER |
The default average producing decimal ranks surprises people coming from SQL.
Ranking within groups
df['rank_in_region'] = (df.groupby('region')['revenue']
.rank(method='dense', ascending=False))
top3 = df[df['rank_in_region'] <= 3]The pandas equivalent of RANK() OVER (PARTITION BY region ORDER BY revenue DESC).
Sorting the index
df.sort_index()
df = df.reset_index(drop=True) # renumber 0..n after filteringdrop=True discards the old index rather than keeping it as a column — usually what you want after filtering.