pd.expanding_mean已弃用：咨询替代方案及弃用原因

Hey there! Those FutureWarnings are pandas' way of nudging you away from an old API toward better, more consistent tools. Let's cover the replacements for pd.expanding_mean and why it's being phased out.

Alternatives to pd.expanding_mean #

For pandas Series or DataFrames #

The modern, preferred replacement is the chainable .expanding().mean() method. This fits seamlessly with pandas' current window operation pattern:

import pandas as pd

# Sample data
df = pd.DataFrame({'values': [10, 20, 30, 40, 50]})

# Deprecated approach
# pd.expanding_mean(df['values'])

# New, supported approach
df['values'].expanding().mean()

This returns the same expanding mean result but uses pandas' standardized window API, which works with other aggregations too (like .sum(), .std()) if you need them later.

For NumPy ndarrays #

If you're working directly with numpy arrays, you have two solid options:

1. Wrap the array in a pandas Series first, then use the expanding mean method:

   import numpy as np
   arr = np.array([10, 20, 30, 40, 50])
   expanding_mean = pd.Series(arr).expanding().mean().to_numpy()
   

2. Use numpy's built-in functions to calculate it manually (great if you want to avoid pandas entirely):

   cumulative_sums = np.cumsum(arr)
   counts = np.arange(1, len(arr) + 1)
   expanding_mean = cumulative_sums / counts
   

Why pd.expanding_mean Was Deprecated #

There are three key reasons pandas decided to phase out this function:

• API Consistency: Pandas has shifted to a modular window API (.expanding(), .rolling(), .ewm()) that returns a window object. You then apply aggregation functions to this object, which makes the code more readable and consistent across different window operations. pd.expanding_mean was a one-off function that broke this pattern.
• Reduced Redundancy: Instead of having separate functions like pd.expanding_mean, pd.expanding_sum, pd.expanding_std, etc., the new API lets you use a single .expanding() call with any aggregation method. This cleans up the pandas namespace and makes it easier to learn.
• Streamlined Support for ndarrays: Pandas' core focus is on its own Series and DataFrame structures. Direct support for ndarrays in the old expanding functions was removed to simplify the codebase. Wrapping an array in a Series (as shown above) is the intended way to handle this now, keeping things aligned with pandas' design principles.

内容的提问来源于stack exchange，提问作者Daniel Chepenko