多层索引DataFrame用iloc报Too many indexers？loc正常原因解析

Great question—this is a super common gotcha when working with pandas MultiIndexes, and it all boils down to how loc and iloc are designed to handle indexing differently. Let’s break this down step by step:

First, let’s clarify your column structure #

From your code and description, it looks like your DataFrame has a 3-level MultiIndex for columns:

• Level 0: Fixed to ['Sale']
• Level 1: Year values (1950, 1951, 1952, etc.)
• Level 2: Week labels (Week1, Week2, Week3, etc.)

Why loc + pd.IndexSlice works perfectly #

loc is a label-based indexer, built explicitly to handle MultiIndex hierarchies. Here’s what’s happening:

• pd.IndexSlice is just a convenience tool to write clean, multi-level slicing logic. When you use idx[:,1950:1952,'Week1':'Week3'], each slice corresponds directly to one level of your column MultiIndex:
  • The first : matches all labels in level 0 (so just Sale)
  • 1950:1952 targets the year labels in level 1
  • Week1:Week3 targets the week labels in level 2
• Since this syntax aligns perfectly with the number of levels in your MultiIndex, loc can parse it without any issues.

Why iloc throws "Too many indexers" #

iloc is a position-based indexer, and it operates differently—no matter how many levels your columns have, iloc sees columns as a single, flat list of integer positions. Here’s the problem with your code:

• iloc only accepts up to two dimensions of indexing (one for rows, one for columns). Your idx[:,:4,:4] tries to pass three dimensions to the column argument, which iloc can’t interpret. It doesn’t understand MultiIndex hierarchies, so it sees this as "too many indexers" for a single dimension (columns).

How to replicate your logic with iloc #

If you need to use integer positions instead of labels, you have to treat columns as a flat list. Here are two ways to do it:

1. Simple flat slicing (if you know the exact column positions) #

If the columns you want are the first 4 in the flat list, just use a single slice:

df.iloc[:, :4]

2. Target specific MultiIndex level positions #

If you need to filter based on positions within each level (e.g., first 4 years and first 4 weeks), you can create a boolean mask to select the right columns:

# Get positions for first 4 years in level 1 and first 4 weeks in level 2
year_pos_mask = df.columns.get_level_values(1).isin(df.columns.levels[1][:4])
week_pos_mask = df.columns.get_level_values(2).isin(df.columns.levels[2][:4])

# Combine masks and slice with iloc
df.iloc[:, year_pos_mask & week_pos_mask]

内容的提问来源于stack exchange，提问作者Vishesh