3. DataFrame: Label-Based & Positional Access

3. Label-Based & Positional Access

Access cells, rows, or slices by label or position.

• df.at[row_label, col_label] → Fast label-based scalar access (single cell).
• df.iat[row_index, col_index] → Fast position-based scalar access.
• df.loc[] → Select rows/columns by label; supports slices, conditions, or lists.
• df.iloc[] → Select rows/columns by integer position.

Key: at/iat → single cell (fast). loc/iloc → slices or multiple rows/columns.

Core Access Methods Overview

We'll use the Titanic dataset for examples:

python

import pandas as pd
import seaborn as sns
df = sns.load_dataset('titanic').set_index('class')

1. df.at[] - Label-Based Scalar Access

DataFrame.at[row_label, column_label]

Key Features

• Fastest method for single cell access
• Works only with existing labels
• Returns a single value (not a Series/DataFrame)

Examples

python

# Get fare for First class passenger at index 1
fare = df.at['First', 'fare']  # Returns scalar value

# Update age of Third class passenger at index 3
df.at['Third', 'age'] = 22  # In-place modification

Professional Usage

python

# Safe access with try-except
try:
    value = df.at['Nonexistent', 'column']
except KeyError:
    value = None

# Bulk updates with .at
for cls in ['First', 'Second']:
    df.at[cls, 'fare'] *= 1.1  # 10% fare increase

2. df.iat[] - Position-Based Scalar Access

Syntax

python

DataFrame.iat[row_position, column_position]

Key Features

• Position-based (zero-indexed)
• Faster than iloc for single cells
• No label checking - pure integer positions

Examples

python

# Get value at row 0, column 2
value = df.iat[0, 2]  # Returns age of first passenger

# Update value at row 5, column 1
df.iat[5, 1] = 'female'  # Change sex of 6th passenger

Performance Comparison

python

%timeit df.at['First', 'fare']  # ~1.5 μs
%timeit df.iat[0, 3]            # ~1.3 μs
%timeit df.loc['First', 'fare'] # ~5 μs

3. df.loc[] - Label-Based Selection

Syntax

python

DataFrame.loc[row_selection, column_selection]

Selection Types

1. Single Label

python

Copy

df.loc['First', 'age']  # Series of ages for First class

2. List of Labels

python

Copy

df.loc[['First', 'Third'], ['age', 'survived']]

3. Slice with Labels

python

Copy

df.loc['First':'Second', 'age':'fare']  # Inclusive!

4. Boolean Masks

python

Copy

df.loc[df['age'] > 30, 'survived']  # Survival status for >30

Advanced Usage

python

# Conditional assignment
df.loc[df['age'] < 18, 'category'] = 'child'

# Cross-section with xs
df.loc[df.index.drop_duplicates(), :]  # Remove duplicate indices

4. df.iloc[] - Position-Based Selection

Syntax

python

DataFrame.iloc[row_positions, column_positions]

Selection Types

1. Integer Position

python

Copy

df.iloc[5, 2]  # 6th row, 3rd column

2. Lists/Slices

python

Copy

df.iloc[10:15, [0, 2, 4]]  # Rows 10-14, specific columns

3. Boolean Masks

python

Copy

df.iloc[(df['age'].values > 30).nonzero()[0]]

Performance Tip

python

# Faster than loc for position-based operations
%timeit df.iloc[100:200, 3:5]  # ~50 μs
%timeit df.loc[df.index[100:200], df.columns[3:5]]  # ~150 μs

Professional Access Patterns

1. Safe Access Wrapper

python

def safe_access(df, row, col, accessor='loc', default=None):
    try:
        if accessor == 'loc':
            return df.loc[row, col]
        elif accessor == 'iloc':
            return df.iloc[row, col]
        elif accessor == 'at':
            return df.at[row, col]
        elif accessor == 'iat':
            return df.iat[row, col]
    except (KeyError, IndexError):
        return default

2. Chained Operation Protection

python

# Dangerous (may cause SettingWithCopyWarning)
df[df['age'] > 30]['fare'] = 100

# Safe using loc
df.loc[df['age'] > 30, 'fare'] = 100

3. Multi-Index Access

python

# Create MultiIndex example
multi_df = df.set_index(['class', 'who'])

# Cross-section access
multi_df.loc[('First', 'man'), :]

# Partial label selection
multi_df.loc[('First', slice(None)), 'age']

Access Method Selection Guide

Performance Optimization

1. Pre-filter columns before row operations:

python

Copy

cols = ['age', 'fare']
df.loc[df['survived'] == 1, cols]

2. Use numpy arrays for complex operations:

python

Copy

mask = (df['age'].values > 30) & (df['sex'].values == 'male')
df.iloc[mask.nonzero()[0]]

3. Avoid chained indexing:

python

Copy

# Bad
df['age'][df['sex'] == 'male'] = 30

# Good
df.loc[df['sex'] == 'male', 'age'] = 30

This comprehensive guide provides professional patterns for efficient DataFrame access in real-world data analysis workflows. The key is matching the access method to your specific selection needs while maintaining code clarity and performance.