Pandas技术疑问：使用iloc检索数据为何与输入索引不匹配？

Great question! Let's clear up the confusion between loc and iloc—this is a super common pitfall when working with pandas DataFrames.

The Core Difference: Label vs. Position Indexing #

The key issue here boils down to what each method uses to fetch rows:

• loc is label-based: it references the actual index labels (the values in your DataFrame's index column, like 1, 2, 50, 9350 in your example).
• iloc is position-based: it looks at the physical order of rows in the DataFrame, starting from 0 (so the first row is position 0, second is 1, and so on).

Why the Mismatch Occurs #

When you run data.loc[data.contributor_message == '...'].index, you're collecting label indices, not physical row positions. If your DataFrame uses the default continuous integer index (0, 1, 2, ...), labels and positions line up perfectly—so iloc[[1,2,50]] works as expected. But if your index has been modified (e.g., you deleted rows, filtered the DataFrame without resetting the index, or set a custom non-continuous index), the labels no longer match the physical positions.

For example:
Suppose you deleted row 3 from your DataFrame. Your index might now look like [0,1,2,4,5,...]. If you use loc to get label 4, that’s the 4th row (physical position 3), but iloc[4] would grab the 5th row (physical position 4, label 5)—two completely different rows!

The Fix: Use the Right Tool for the Job #

Since you’re working with label indices from loc, stick with loc to retrieve the matching rows:

# Get the matching label indices
matching_labels = data.loc[data.contributor_message == 'I support this proposal because...'].index
# Retrieve the rows using loc
matching_rows = data.loc[matching_labels]

Or even simpler, skip the index step entirely:

matching_rows = data[data.contributor_message == 'I support this proposal because...']

Only use iloc when you need to reference rows by their physical position (e.g., "give me the first 5 rows" with iloc[:5], or "give me the 10th row" with iloc[9]).

内容的提问来源于stack exchange，提问作者RyanKao