Pandas loc versus iloc indexing

What's really being asked

This checks day-to-day pandas competence and attention to the inclusive-versus-exclusive slicing rule that trips up many engineers. It signals whether you write correct data-wrangling code.

The full answer

The loc accessor selects by label: you pass index values and column names. Its slice is inclusive of both the start and the stop label, so df.loc['a':'c'] returns rows a, b, and c. The iloc accessor selects by integer position from 0 to length minus 1, following standard Python slicing where the stop is excluded, so df.iloc[0:3] returns positions 0, 1, and 2. When the DataFrame has a default integer RangeIndex the two can look similar, but the inclusive-versus-exclusive boundary still differs.

The mistakes people make

Saying both are interchangeable. Claiming both slices are exclusive, or both inclusive. Believing loc always needs strings; loc works on any label type including integers when the index is integer-labeled. Forgetting that mixing them silently produces wrong rows rather than always erroring.

What usually comes next

What does df.loc index when the index itself is integers but unsorted. How do boolean masks interact with loc. What is the difference between df['col'] and df.loc[:, 'col'].

A concrete example

Suppose df has a string index ['x', 'y', 'z']. Calling df.loc['x'] returns the first row by label and works fine. Calling df.iloc['x'] raises a TypeError because iloc demands integer positions, not labels. Conversely df.iloc[0] returns the first row, while df.loc[0] raises a KeyError because no row is labeled 0. This contrast is exactly the example interviewers want, showing each accessor fails when given the other's input type.