Indexing and Selecting Data (loc, iloc)

Why selection matters

Most Pandas bugs come from:

• Selecting the wrong rows
• Accidentally getting a copy instead of a view
• Confusing label-based vs position-based indexing

Pandas provides two primary tools:

• loc → label-based
• iloc → position-based

From the book

McKinney’s advice: prefer loc over plain [] indexing whenever your index could contain integers. Regular [] indexing treats integers as labels if the index itself holds integers, so behavior silently changes depending on your data. loc always means “by label” and iloc always means “by position” — no ambiguity either way.

Setup example DataFrame

Sample DataFrame

import pandas as pd
 
df = pd.DataFrame(
    {
        "name": ["Asha", "Ravi", "Meera", "Zoya"],
        "age": [23, 28, 26, 31],
        "city": ["Pune", "Delhi", "Delhi", "Pune"],
        "score": [88, 91, 77, 95],
    },
    index=["u1", "u2", "u3", "u4"],
)
 
print(df)

Selecting columns

One column

print(df["age"])  # returns a Series

Multiple columns

print(df[["name", "score"]])

Selecting rows with loc (labels)

loc by label

print(df.loc["u2"])          # one row (Series)
print(df.loc[["u1", "u4"]]) # multiple rows (DataFrame)

Selecting rows + columns together:

loc rows + cols

print(df.loc[["u1", "u3"], ["name", "score"]])

Selecting rows with iloc (positions)

iloc by position

print(df.iloc[0])      # first row
print(df.iloc[0:2])    # first two rows
print(df.iloc[:, 0:2]) # first two cols

Boolean filtering (most common)

Filter rows

filtered = df[df["city"] == "Delhi"]
print(filtered)

Combine conditions:

Multiple conditions

filtered = df[(df["city"] == "Pune") & (df["score"] >= 90)]
print(filtered)

Safer assignment after filtering

When you filter and then assign, prefer .loc:

Safe assignment with loc

df.loc[df["city"] == "Delhi", "is_delhi"] = True
 
df["is_delhi"] = df["is_delhi"].fillna(False)
print(df)

Note on & and |

In Pandas boolean filters:

• Use & for AND, | for OR
• Always wrap sub-conditions in parentheses

This avoids operator precedence bugs.

Visualize it

loc reads coordinates as labels (“row u2, column age”); iloc reads them as positions (“row 1, column 1”) — same grid, two different ways to point at a cell.

diagram loc vs iloc mermaid

  flowchart LR
  DF["DataFrame grid"] --> L["df.loc['u2', 'age']  (by label)"]
  DF --> I["df.iloc[1, 1]  (by position)"]
  L --> V["Same cell"]
  I --> V

loc addresses cells by label, iloc addresses the same cells by integer position.

sketch Label-based vs position-based selection p5.js Pause Restart

loc and iloc point at the exact same cell using two different coordinate systems.

🧪 Try It Yourself

Exercise 1 – Select With loc

pch.exerciseTag pch.exerciseTitle

Exercise 2 – Select With iloc

pch.exerciseTag pch.exerciseTitle

Exercise 3 – Rows + Columns Together

pch.exerciseTag pch.exerciseTitle

Next

Selection often needs a differently shaped index first — see Reindexing and Data Alignment for how pandas rearranges and aligns data by label.