Control decimal precision
Normalize measured or monetary-like floating-point columns before display or downstream calculations.
Polars · Module
.round
Rounds floating-point expression values to a specified number of decimal places. The default half_to_even mode resolves ties by choosing the nearest even value.
How it works
Expr.round returns an expression that rounds floating-point values to a chosen number of decimal places. It uses the selected strategy for tie handling or, with to_zero, truncates toward zero.
Syntax
Expr.round(decimals: int = 0, mode: RoundMode = 'half_to_even') → ExprParameters
| Name | Type | Description |
|---|---|---|
decimals | int = 0 | Number of decimals to round by. |
mode | RoundMode = 'half_to_even' | Rounding strategy; one of half_to_even, half_away_from_zero, or to_zero. |
Returns
Expr
Key points
- The expression rounds floating-point values to the number of decimal places specified by decimals.
- The result has no more than the requested number of decimal places; zero decimals produces whole-number precision.
- The mode parameter selects how rounding is performed.
- Both half_to_even and half_away_from_zero first choose the nearest rounded value.
- half_to_even resolves exact halfway cases by selecting the even result.
- half_away_from_zero resolves exact halfway cases in the direction away from zero.
- to_zero discards the remaining fractional portion toward zero at the requested precision.
Common mistakes
- Choose to_zero when truncation toward zero is intended; the documentation identifies that behavior as equivalent to truncate().
Example: Round an expression
import polars as pl
df = pl.DataFrame({"price": [1.234, 2.345, 2.5]})
result = df.select(pl.col("price").round(2))
print(result)Rounds the selected floating-point column to two decimal places using the default half-to-even strategy.
When to use it
Choose a rounding policy
Apply predictable tie-breaking when aggregating or presenting numeric results.
Truncate toward zero
Remove fractional precision without nearest-value rounding.
Interview questions
QWhat does Expr.round do, and what is its default precision?
It rounds the underlying floating-point data by the requested number of decimal digits and returns an Expr. With decimals=0, values are rounded to whole-number precision.
Follow-up: What happens for an exact tie under the default mode? The nearest even rounded value is selected, such as 0.5 to 0 and 1.5 to 2.
QHow does half_to_even differ from half_away_from_zero?
half_to_even rounds to the nearest value and resolves ties using the nearest even value. This is the default strategy and is also called banker’s rounding.
Follow-up: Why use it by default? The documentation says it tends to minimize cumulative rounding bias.
QWhen should the to_zero mode be used?
The to_zero mode always truncates toward zero, discarding fractional digits beyond the requested precision. Unlike half-based modes, it is not only a tie-breaking rule; it always determines the chosen rounded value.
Follow-up: Which related operation has equivalent behavior? The documentation states that Expr.round with to_zero is equivalent to truncate().
Related
polars.Series.roundpolars.Expr.truncatepolars.Expr.dt.round