Acceso por etiquetas y por posiciones

Pandas provides two primary accessors for selecting elements from a Series: .loc for label-based access and .iloc for integer position-based access. Understanding which to use and when is critical — using the wrong one on an integer-indexed Series produces confusing results, and Pandas 2.0 deprecated many ambiguous direct bracket usages.

.loc accesses elements by their index label. You can pass a single label (returns a scalar), a list of labels (returns a Series), or a slice of labels (inclusive on both ends, unlike Python slices). If the label does not exist, Pandas raises a KeyError. Use .loc when labels are meaningful (dates, names, IDs).

.iloc accesses elements by their integer position (0-based), completely ignoring the index labels. This behaves like Python list indexing: negative positions count from the end, and slices follow Python's exclusive-stop convention. Use .iloc when you need the 'n-th' element regardless of its label.

The difference matters most for integer-labelled Series. If a Series has index labels [5, 10, 15], then s.loc[5] returns the element with label 5 (the first element) while s.iloc[5] raises an IndexError (only 3 elements exist). Always choose based on whether you mean 'the element labelled X' or 'the X-th element'.

Pass a list to .loc to select multiple elements in any order, potentially with repetitions. The result is a new Series with the same labels as the list you provided. This is similar to fancy indexing in NumPy but aligned on index labels rather than positions. It is commonly used to reorder or subset columns in analytical pipelines.

Unlike Python slices, .loc slices are inclusive on both ends. s.loc['b':'d'] returns elements with labels b, c, and d. This can be surprising if you are used to Python's exclusive-stop convention. Make sure the labels you slice between actually exist in the index, or the slice will return unexpected results silently.

You can pass a boolean Series to .loc to filter elements. The boolean Series must have the same index as the target Series. This is the Pandas equivalent of NumPy boolean masking and supports compound conditions with & and |. It is the standard way to filter rows in DataFrames too.

Both .loc and .iloc can appear on the left side of an assignment to modify values in place. This is the correct pattern — using chained indexing like s['a'] = 99 may trigger a SettingWithCopyWarning in DataFrames. For Series, direct label access is generally safe, but .loc is explicit and recommended.

Direct bracket notation s['label'] works for label access and is equivalent to s.loc['label'] for string indices. For integer indices the behaviour is ambiguous and was changed in Pandas 2.0 — it now always means label access. For clarity and future-proof code, always use .loc or .iloc explicitly rather than relying on bracket disambiguation.

When a Series has a DatetimeIndex, .loc accepts partial date strings for convenient range selection. s.loc['2023'] returns all elements from 2023, and s.loc['2023-01':'2023-06'] returns the first half of 2023. This is one of Pandas' most ergonomic features for time series work.

.at[label] and .iat[position] are optimised scalar accessors — faster than .loc and .iloc when accessing a single element in a tight loop, because they skip the overhead of returning a Series object. They cannot select multiple elements. Use them when you need maximum performance for repeated single-element lookups.

Test your understanding of .loc and .iloc from this lesson.

In this lesson you learned: .loc accesses elements by index label with inclusive slice stops, .iloc accesses elements by integer position with exclusive slice stops like Python lists, and .at and .iat are faster scalar accessors for single-element lookups in loops. Next up we explore vectorized operations on Series and how Pandas handles index alignment automatically.