fillna方法中method='pad'与method='ffill'的差异及优选性探讨

Great question—you’re totally right to notice they seem identical, because they actually are fully equivalent! Let’s break this down clearly:

• Core Relationship: pad is just an alias for ffill in pandas. That means under the hood, calling df.fillna(method='pad') runs exactly the same code as df.fillna(method='ffill').
• Why Two Names?: It’s mostly about naming conventions and compatibility. ffill is short for forward fill, which makes its behavior immediately obvious (fill missing values with the previous non-null entry). pad leans into the literal "pad" (fill in gaps) terminology, which some developers might prefer for brevity.
• Behavior Check: Here’s a quick snippet to confirm they produce identical results:

  import pandas as pd
  import numpy as np
  
  df = pd.DataFrame({'values': [2, np.nan, 5, np.nan, 9]})
  # Both lines will fill NaNs with 2 and 5 respectively
  result_pad = df.fillna(method='pad')
  result_ffill = df.fillna(method='ffill')
  print(result_pad.equals(result_ffill))  # Output: True
  

Preferred Method? #

While there’s no functional difference, ffill is generally the more popular choice in modern pandas code. Its explicit "forward fill" name makes the code more readable for anyone reviewing it—no need to parse what "pad" refers to in this context. That said, if you encounter pad in legacy code, you can be confident it’s doing the exact same thing as ffill.

For completeness, the same alias relationship applies to backward filling: backfill is the alias for bfill, and bfill is the more commonly used option.

内容的提问来源于stack exchange，提问作者jhourback