Data Structures

The containers the operator stores work in. Python’s built-in types cover most everyday needs; collections, heapq, queue, array, and struct cover the specialized cases. Picking the right structure up front is usually the difference between a recipe that runs in seconds and one that times out on real data.

For the primitive and composite types themselves, see Types. For the algorithms that run over them, see Algorithms.

Built-in

The everyday types are documented under Types.

list

Mutable, ordered, indexable sequence.

list

tuple

Immutable, ordered sequence.

tuple

dict

Insertion-ordered key-value map.

dict

set

Mutable collection of unique hashable values.

set

frozenset

Immutable, hashable set.

frozenset

range

Lazy integer sequence.

range

Records

Fixed-shape data with named fields.

Records

namedtuple, @dataclass, TypedDict. The three ways to give a fixed-shape value a name.

Records

Specialized containers

The collections module covers the gaps in the built-ins.

deque

Double-ended queue. O(1) at both ends.

deque

defaultdict

Dict that auto-initialises missing keys with a factory.

defaultdict

Counter

Multiset / frequency map.

Counter

OrderedDict

Ordered dict with move_to_end and order-aware equality.

OrderedDict

ChainMap

Layered dict lookup, leftmost first.

ChainMap

Heap and queues

Heap

heapq: binary min-heap on a list. Priority queues and top-N extraction.

Heap

Queues

queue.Queue, multiprocessing.Queue, asyncio.Queue for producer-consumer.

Queues

Binary buffers

struct

Pack and unpack fixed-layout binary records.

struct

memoryview

Zero-copy view over a bytes-like object.

memoryview

BytesIO / StringIO

In-memory file-like objects.

BytesIO and StringIO

Bitwise integers

Bit operations on arbitrary-precision int.

Bitwise integers

Tabular (third-party)

Pandas containers; third-party (pip install pandas) but the operator’s default for any tabular workload that fits in memory.

Series

pandas.Series: 1-D labelled array. Index, values, dtype.

Series

DataFrame

pandas.DataFrame: 2-D labelled table. Rows of records, columns of Series.

DataFrame

The wider ecosystem; reach for these when pandas does not fit the workload.

Library	When to reach for it
pandas	The default. Eager, single-threaded, in-memory. Largest documentation and recipe surface.
polars	Faster than pandas (Rust + Arrow); supports a lazy query planner; first choice for any workload that pegs pandas on a single machine.
pyarrow	Apache Arrow in Python. The columnar interchange format (zero-copy across pandas, polars, duckdb, Parquet, Spark); also a standalone tabular layer.
duckdb	In-process SQL engine that reads Arrow / pandas / polars directly. The right tool when the question reads as SQL.
ibis	Backend-portable DataFrame API; one Python expression compiles to duckdb, BigQuery, Postgres, Snowflake, polars.
modin	Drop-in import modin.pandas as pd replacement that parallelises across cores and a Ray / Dask cluster.
dask.dataframe	Partitioned pandas across a cluster; lazy graph; larger-than-memory workloads on commodity boxes.
vaex	Memory-mapped, out-of-core; billion-row exploration on a laptop.
pyspark.sql.DataFrame	Apache Spark from Python; cluster-scale; the production choice when the data exceeds a single host.

Choosing

Choosing

The operator’s cheat-sheet for picking the right container.

Choosing

References

• Types for the primitive and composite types.

• Algorithms for the algorithms over them.

• Syntax for slicing and comprehensions.

• Python Standard Library, Data Types — the reference for collections, heapq, queue, array, struct, io, copy, enum, types.

• Time complexity wiki — CPython container performance.