Column-wise ordering
Compare observations within each feature by ordering values from smallest to largest or vice versa.
Pandas · Module
df.rank
DataFrame.rank assigns numerical ranks along columns by default or across rows when requested. It supports configurable tie, missing-value, direction, numeric-column, and percentile behavior.
How it works
DataFrame.rank computes numerical ranks independently along each column by default, or across each row when axis=1. It lets you choose tie handling, NaN placement, sort direction, numeric-column filtering, and percentile output while preserving the caller's Series or DataFrame shape.
Syntax
DataFrame.rank(axis=0, method='average', numeric_only=False, na_option='keep', ascending=True, pct=False)Parameters
| Name | Type | Description |
|---|---|---|
axis | {0 or 'index', 1 or 'columns'} | Axis used for ranking: 0 or 'index' ranks down columns; 1 or 'columns' ranks across rows. Default: 0. |
method | {'average', 'min', 'max', 'first', 'dense'} | Tie-ranking strategy: 'average', 'min', 'max', 'first', or 'dense'. Default: 'average'. |
numeric_only | bool | For DataFrames, rank only numeric columns when True. Default: False. |
na_option | {'keep', 'top', 'bottom'} | Missing-value policy: 'keep', 'top', or 'bottom'. Default: 'keep'. |
ascending | bool | Rank in ascending order when True; default: True. |
pct | bool | Return rankings in percentile form when True. Default: False. |
Returns
same type as caller: a Series or DataFrame containing ranks.
Key points
- The method produces numerical ranks along either DataFrame axis.
- axis accepts 0 or 'index' for column-wise ranking, and 1 or 'columns' for row-wise ranking.
- method controls how tied values are ranked: average, min, max, first, or dense.
- ascending=True gives smaller values lower ranks; set it to False to rank larger values first.
- na_option determines whether missing values remain missing or receive the lowest or highest rank.
- The result keeps the same general type as the caller: a DataFrame for DataFrame.rank.
Common mistakes
- The default tie strategy is average, so equal values can receive fractional ranks; choose min, max, first, or dense when a different policy is needed.
- numeric_only=True excludes nonnumeric columns; the default is False.
- pct=True returns percentile-style ranks rather than ordinary ordinal ranks.
Example: Rank rows with ties
import pandas as pd
scores = pd.DataFrame({"alice": [80, 95], "bob": [90, 95]})
print(scores.rank(axis=1, method="dense", ascending=False))Ranks each row from highest score to lowest, assigning equal scores the same dense rank.
When to use it
Row-wise comparisons
Create per-row leaderboards across several measurements with axis=1.
Dense tie groups
Convert tied values into contiguous group numbers using method='dense'.
Percentile rankings
Express a value's relative rank as a percentile with pct=True.
Interview questions
QWhat does DataFrame.rank do, and how are ties handled by default?
DataFrame.rank assigns numerical ranks from 1 through n along the chosen axis. With the default method='average', tied values receive the mean of the ranks they occupy.
Follow-up: How would you rank values within rows and make ties share consecutive group numbers? Use axis=1 and method='dense'.
QHow does axis change DataFrame.rank's behavior?
axis=0, also called 'index', ranks down each column. axis=1, also called 'columns', ranks across each row.
Follow-up: How can missing values be placed at the beginning? Set na_option='top'; use 'bottom' to place them at the highest ranks.
QWhen would you choose method='first' versus method='dense'?
The method choices are 'average', 'min', 'max', 'first', and 'dense'. 'first' breaks ties by their order of appearance, while 'dense' increases the rank by one between distinct value groups.
Follow-up: What does pct=True change? It returns rankings in percentile form instead of the usual integer-like rank scale.
Related
Series.rankDataFrameGroupBy.rankpandas.core.window.expanding.Expanding.rank