Sorting and Ranking

Two different questions: sort vs rank

• Sorting rearranges rows so they appear in order.
• Ranking assigns each value a position number (1st, 2nd, 3rd…) without rearranging anything — the rank becomes a new column or Series aligned to the original order.

Both are everyday analytics needs: sorting a leaderboard, ranking students by score.

Sorting by index: sort_index

sort_index

import pandas as pd
import numpy as np
 
obj = pd.Series([4, 7, -3, 2, np.nan], index=["d", "a", "c", "b", "e"])
print(obj.sort_index())  # sorted lexicographically by label

Sorting by value: sort_values

sort_values

print(obj.sort_values())

Missing values sort to the end by default. Use na_position="first" to flip that:

Controlling where NaN lands

print(obj.sort_values(na_position="first"))

Sorting a DataFrame by one or more columns

Sort a DataFrame by column values

frame = pd.DataFrame({"b": [4, 7, -3, 2], "a": [0, 1, 0, 1]})
 
print(frame.sort_values("b"))          # single column
print(frame.sort_values(["a", "b"]))   # tie-break: sort by "a" first, then "b"

Ranking: rank()

rank() assigns each value a position from 1 through the number of valid points, starting at the lowest value. Ties are handled by averaging their ranks by default:

rank() with default tie-breaking

s = pd.Series([7, -5, 7, 4, 2, 0, 4])
print(s.rank())

Both 7s tie for ranks 6 and 7, so they each get the average, 6.5. Other tie-break rules are available via method:

Tie-breaking methods

print(s.rank(method="first"))  # ties broken by order of appearance
print(s.rank(method="min"))    # ties get the lowest rank in the group
print(s.rank(ascending=False)) # rank from highest to lowest instead

method	What it does
"average"	Default — tied values share the mean rank
"min"	Tied values all get the lowest rank in the group
"max"	Tied values all get the highest rank in the group
"first"	Ties broken by the order values appear in the data
"dense"	Like "min", but ranks always increase by exactly 1 between groups

Common pitfalls

• sort_values() on a Series with NaN puts them last — easy to miss if you don’t scroll to the bottom of a big result.
• rank()’s default "average" method can produce fractional ranks (6.5) — if you need whole-number ranks for display, use method="first" or method="dense".
• Sorting returns a new object; it doesn’t sort in place unless you pass inplace=True (and even then, prefer reassigning — it’s clearer).

Visualize it

diagram Sort vs rank mermaid

  flowchart LR
  A["Unsorted values"] -->|"sort_values()"| B["Rows physically reordered"]
  A -->|"rank()"| C["Same row order,
each value gets a rank number"]

sort_values rearranges rows into order; rank keeps the original row order but attaches a position number to each value.

sketch Bars reordering by sort_values p5.js Pause Restart

The faded bars show the original order; the solid bars show the same values after sort_values() rearranges them ascending.

From the book

McKinney notes that rank()’s default tie-breaking method, "average", means two tied values each receive the mean of the ranks they would have occupied. If you’ve ever wondered why a rank column has decimals like 6.5, this is why — switch to method="first" or method="min" if you need whole numbers.

🧪 Try It Yourself

Exercise 1 – Sort a Series by Value

pch.exerciseTag pch.exerciseTitle

Exercise 2 – Put Missing Values First

pch.exerciseTag pch.exerciseTitle

Exercise 3 – Rank Values With Averaged Ties

pch.exerciseTag pch.exerciseTitle

Next

Sorting rearranges rows you already have. Applying Functions (apply, map, applymap) covers transforming the values inside those rows.