Top-k minimum selection
Find the lowest-valued records in a DataFrame without sorting the entire table.
Pandas · Module
df.nsmallest
Returns the first n DataFrame rows with the smallest values in one or more columns, ordered ascending. Other columns are included but do not influence ranking.
How it works
Call nsmallest with a row count and one or more ordering columns. It returns the selected rows in ascending order, retaining every DataFrame column; when values tie, keep controls which occurrences are retained.
Syntax
DataFrame.nsmallest(n, columns, keep='first')Parameters
| Name | Type | Description |
|---|---|---|
n | int | Number of items to retrieve. |
columns | list or str | Column name or names to order by. |
keep | {'first', 'last', 'all'}, default 'first' | Tie behavior: 'first' keeps the first occurrence, 'last' the last occurrence, and 'all' every tie at the boundary; defaults to 'first'. |
Returns
DataFrame with the first n rows ordered by columns in ascending order.
Key points
- The method selects the smallest n rows and orders the result from low to high by the chosen columns.
- All DataFrame columns remain available in the result, even though only the listed columns determine ranking.
- It is a concise, faster alternative to sorting ascending and taking the first n rows.
- The n argument specifies how many rows to request.
- The columns argument accepts either one column name or multiple names for ordering.
- The default keep='first' preserves the earliest occurrence among tied values.
Common mistakes
- Using keep='all' can return more than n rows when the boundary value has ties.
- With keep='last', duplicate values are selected from the reverse order of appearance.
Example: Select rows with the smallest values
import pandas as pd
df = pd.DataFrame({
"city": ["A", "B", "C", "D"],
"population": [120, 80, 80, 200],
"gdp": [4, 7, 3, 9],
})
result = df.nsmallest(2, ["population", "gdp"], keep="first")
print(result)Selects the two rows with the smallest population, breaking ties by GDP.
When to use it
Multi-column ordering
Rank by a primary metric and use additional columns to break ties.
Tie-preserving selection
Retain every record tied at the cutoff rather than forcing exactly n rows.
Interview questions
QWhat does DataFrame.nsmallest do?
DataFrame.nsmallest(n, columns, keep='first') returns the first n rows ordered by the specified columns in ascending order. It includes unspecified columns in the result, but those columns do not affect ordering.
Follow-up: How would you reproduce it without nsmallest? Use df.sort_values(columns, ascending=True).head(n), although the documentation notes nsmallest is more performant.
QHow do n and columns control selection?
n is the number of items to retrieve, while columns is a column name or a list of names used for ordering. Multiple columns apply successive ordering criteria, so later columns resolve ties from earlier ones.
Follow-up: What happens if the ordering columns are omitted? The call is invalid because both n and columns are required positional arguments in the documented signature.
QHow does keep affect ties?
keep determines duplicate handling: first keeps the first occurrence, last keeps the last occurrence, and all keeps every tie at the boundary. Therefore keep='all' may produce more than n rows.
Follow-up: Why might keep='all' be useful? It preserves every row tied at the cutoff instead of arbitrarily dropping tied rows.
Related
DataFrame.nlargestDataFrame.sort_valuesDataFrame.headSeries.nsmallest