Memoization

Avoids expensive recomputation of pure functions by caching their results.

Web development: caching expensive database query results in a view function; Data processing: memoizing recursive Fibonacci calculation; Scientific computing: caching intermediate results in iterative algorithms

Python standard library, web frameworks like Flask and Django, data science pipelines

@lru_cache(maxsize=max_size)

@lru_cache()

max_size: int ≥ 0, maximum number of cached entries

• functools.wraps
• recursive functions
• pure functions

• Using a custom dictionary cache (more control but manual management)
• using @cache (Python 3.9+ simpler LRU cache without maxsize)
• using lru_cache without maxsize (unbounded cache)

Applying @lru_cache to a function with mutable arguments (cause: unhashable types; consequence: TypeError); forgetting to import lru_cache from functools (cause: missing import; consequence: NameError); setting maxsize=0 (cause: misunderstanding; consequence: no caching)

@cache (unbounded LRU cache introduced in Python 3.9); @staticmethod (decorator for utility functions); manual memoization with dict (more verbose)

Coming from Java: may attempt to use static fields for caching → Python's lru_cache is function-specific and thread-safe in CPython due to GIL but not guaranteed across interpreters; Coming from C++: may try to use std::unordered_map directly → lru_cache provides automatic eviction policy

• @cache
• @staticmethod
• @property
• manual dict memoization

Do not use on functions with side effects or non-deterministic output; caching consumes memory proportional to maxsize and can cause memory pressure; the cache key is based on argument values; unhashable arguments raise TypeError

Reduces latency and CPU usage for repeated pure function calls, enabling faster response times in web services and scientific computations.

Like a librarian who keeps the most recently requested books on the front desk for quick access.

In Python 3.8+, lru_cache returns a wrapper with a cache_info() function for statistics.

Consider using @functools.cache for unbounded caching or implementing a TTL-based cache for time-sensitive data.