How to Fill Missing Data with a Rolling Weighted Average in Pandas

What will you learn?

In this comprehensive tutorial, you will delve into the world of handling missing data by leveraging a rolling weighted average in pandas. By mastering this technique, you’ll enhance your data manipulation skills, ensuring accurate analyses even in the presence of data gaps.

Introduction to the Problem and Solution

Encountering missing values while working with time series or sequential data is a common challenge in Python. These gaps can disrupt analytical insights if not addressed effectively. One powerful approach to mitigate this issue is by filling these voids with estimates derived from nearby values using a rolling weighted average method. Unlike simplistic imputation strategies, this method considers the proximity of neighboring data points, offering more precise estimations for the missing values.

To tackle this prevalent issue, we turn to the versatile pandas library known for its robust capabilities in data manipulation and analysis. The solution involves calculating a rolling window around each missing value and computing a weighted average that prioritizes closer observations. This guide equips you with both the practical implementation of this technique and an understanding of its rationale, making it adaptable for diverse scenarios involving incomplete datasets.


import pandas as pd
import numpy as np

# Sample DataFrame creation
df = pd.DataFrame({'Values': [1, np.nan, 3, np.nan, 5]})

# Define function for weighted average filling
def fill_with_weighted_average(series, window_size):
    weights = np.arange(1, window_size + 1)
    return series.rolling(window=window_size*2-1,
                          center=True).apply(lambda x:, weights)/weights.sum(), raw=False)

# Apply function 
df['Filled_Values'] = fill_with_weighted_average(df['Values'], 2)

# Copyright PHD


The provided code snippet demonstrates how to employ a rolling weighted average to fill missing values in pandas:

  • Import essential libraries: pandas for data manipulation and numpy for numerical operations.
  • Create a sample DataFrame df containing hypothetical numeric data with ‘NaN’ representing missing values.
  • The fill_with_weighted_average function takes a pandas Series and the window size around each ‘NaN’ value for averaging.
  • Within the function:
    • Define weights array to emphasize nearby observations.
    • Utilize .rolling() combined with .apply() on the series; temporarily fill NaNs with zero to compute weighted averages accurately.
    • Calculate the weighted average by taking dot product between filled series and weights divided by sum of all weights.
  • Applying fill_with_weighted_average on ‘Values’ column fills missing entries under ‘Filled_Values’.

This approach effectively smooths out fluctuations caused by absent data without imposing arbitrary assumptions on your dataset.

    How does pandas handle rolling computations over NaN values?

    By default, Pandas excludes NaN values during .rolling() operations unless specified otherwise using arguments like min_periods.

    Can I adjust the weighting scheme used here?

    Certainly! Modifying the weights array within fill_with_weighted_average allows customization based on specific requirements.

    What does ‘center=True’ do in .rolling()?

    Setting ‘center=True’ centers the window at each point rather than starting from it�crucial for achieving symmetry around evaluated positions.

    Is there an alternative method if linear weighting doesn’t suit my dataset’s pattern?

    Yes! Consider exponential smoothing or interpolation techniques such as np.interp() based on dataset characteristics.

    Can I apply this method across different columns simultaneously?

    To apply across multiple columns concurrently requires iterating over those columns or implementing modifications enabling vectorized operations across DataFrame directly.

    What about very large datasets? Is this method still efficient?

    For extensive datasets, memory consumption may rise due to rolling windows. Experimenting with larger step sizes or initial down-sampling could alleviate performance concerns.


    Mastering the art of filling missing values through a rolling weighted average grants nuanced control over temporal dynamics during gap-filling processes�a critical aspect when preparing time-series data for profound analysis. Acquiring proficiency in such techniques not only upholds data integrity but also enhances representational accuracy of real-world phenomena encapsulated within our datasets.

    Leave a Comment