Exponentially weighted smoothing
Smooth noisy measurements or financial series while emphasizing recent observations.
Pandas · Module
df.ewm
Creates an exponentially weighted window for DataFrame calculations. Supply a decay definition and then apply an operation such as mean() or std().
How it works
Calling ewm creates an exponentially weighted window over the DataFrame. Choose one decay specification, then call a window operation such as mean(); adjust, missing-value handling, minimum observations, time-based decay, and execution method control the calculation.
Syntax
DataFrame.ewm(com=None, span=None, halflife=None, alpha=None, min_periods=0, adjust=True, ignore_na=False, times=None, method='single')Parameters
| Name | Type | Description |
|---|---|---|
com | float, optional | Center-of-mass decay; requires com >= 0. |
span | float, optional | Span-based decay; requires span >= 1. |
halflife | float, str, timedelta, optional | Half-life decay; accepts a positive numeric or time-like value. |
alpha | float, optional | Direct smoothing factor; 0 < alpha <= 1. |
min_periods | int, default 0 | Minimum observations required; default 0. |
adjust | bool, default True | Whether to use the decaying adjustment factor; default True. |
ignore_na | bool, default False | Whether missing observations affect relative positions; default False. |
times | str, datetime-like, Series, optional | Observation times used for time-aware decay. |
method | {'single', 'table'}, default 'single' | Execution strategy; default 'single'. |
Returns
An exponentially weighted window object supporting exponentially weighted calculations.
Key points
- The method creates an exponentially weighted window for DataFrame calculations.
comexpresses decay using center of mass; its documented relationship is = 1 / (1 + com).spanexpresses decay using span and is converted into the smoothing factor.halflifeexpresses how quickly weights decay by defining a half-life.alphalets callers provide the smoothing factor without converting from another decay representation.- With
adjust=True, early periods are normalized with a decaying adjustment factor.
Common mistakes
- Without
times, supplying none or more than one of the four decay parameters is invalid. - Time-aware windows with
adjust=Falserequirehalflifealone as the decay specification. - With
timesandadjust=True, halflife can be combined with at most one alternative decay parameter.
Example: Exponentially weighted moving mean
import pandas as pd
df = pd.DataFrame({"price": [10, 12, 11, 15]})
result = df.ewm(span=3, adjust=False).mean()
print(result)Creates an exponentially weighted moving mean where span controls decay and earlier values receive progressively less weight.
When to use it
Time-aware decay
Use timestamps when observations are irregularly spaced and decay should reflect elapsed time.
Control weighting behavior
Configure startup behavior and missing-value treatment for a moving calculation.
Interview questions
QWhat does DataFrame.ewm return and how is it used?
ewm returns an exponentially weighted window object rather than the aggregated values immediately. Calling methods such as mean() performs the requested exponentially weighted calculation.
Follow-up: How can decay be specified? Use exactly one of com, span, halflife, or alpha when times is absent; with time-aware calculations, halflife has special combination rules.
QHow can the decay rate be configured?
The decay can be expressed through center of mass, span, half-life, or the smoothing factor directly. The method then applies that weighting to operations such as the exponentially weighted mean.
Follow-up: Which direct smoothing-factor constraint applies? alpha must satisfy the documented range greater than zero and less than or equal to one.
QWhen would you use the times parameter?
times supplies timestamps corresponding to observations, enabling time-based decay through halflife. The permitted decay-parameter combinations depend on whether adjust is true or false.
Follow-up: What restriction applies when adjust=False? halflife must be the only decay-specification parameter when times is provided.
Related
pandas.api.typing.ExponentialMovingWindow.meanpandas.api.typing.ExponentialMovingWindow.stdpandas.api.typing.DataFrameGroupBy.ewmDataFrame.rolling