caching-optimizer

Caching reduces latency in Streamlit applications by storing results of expensive operations. Use @st.cache_data for data loads and transformations, and @st.cache_resource for external connections or ML models. Wrap the function once, and Streamlit automatically reuses cached results on repeated calls with identical inputs.

@st.cache_data caches serializable outputs like dataframes and API responses; @st.cache_resource caches unserializable objects like database connections and loaded models. Use cache_data for computations and transformations; use cache_resource for stateful resources that persist across reruns.

Cache invalidation happens automatically when function parameters differ, ensuring fresh results for new inputs. Implement TTL-based expiration to refresh stale data, use manual cache clearing for on-demand updates, or apply conditional caching logic to balance freshness and performance based on your use case.

Yes. Streamlit's caching handles most built-in types automatically. For complex or custom types, implement custom hash handling to tell the cache how to recognize identical inputs, avoiding unnecessary recomputation while maintaining correctness.

Set size limits when memory is constrained or data is frequently updated; configure TTL for real-time data sources where staleness matters. Without limits, caches grow unbounded; without TTL, old data persists indefinitely. Balance retention against freshness requirements for your analytics or preprocessing pipeline.

Yes. Caching is designed for repeated data loads, API calls, and preprocessing pipelines in iterative analytics workflows. Once cached, subsequent queries return results instantly, improving user experience during exploration without re-fetching or reprocessing.