LAFC-Evict: Learning-Augmented Cache Eviction Dataset

Description

LAFC-Evict is a derived, tabular research dataset of counterfactual supervision labels for learning-augmented cache-eviction research. Each row describes one candidate object considered for eviction at one full-cache-miss decision point, together with engineered state/candidate features and a label describing what would happen to the cache miss rate over a short future horizon if that specific candidate were evicted. It is intended for training and evaluating machine-learning-based cache-eviction policies and for studying disagreement between learned predictors and classical eviction heuristics such as LRU. LAFC-Evict is a generated/derived artifact: it is computed by the dataset author from third-party, CC0-1.0 Wikimedia pageview data, and it does not redistribute any raw Wikimedia pageview rows, raw page titles, or third-party trace data. Upstream Wikimedia data must be cited separately according to its own terms; this release claims authorship only of the derived candidate-row features and labels, not of the underlying popularity signal. The current public release (v0.3) contains 22,356,992 rows across two Parquet configurations, built entirely from the wiki2018 trace family. A smaller, fully synthetic companion dataset, SoroushVahidi/lafc-evict-sample, exists separately as a packaging/format smoke-test fixture and is not a scientific release.

Update Frequency

No fixed schedule. Updated on major dataset releases (see Version history in the dataset documentation); the current version is v0.3, published 2026-08-13.

License

CC0 1.0

Documentation

https://huggingface.co/datasets/SoroushVahidi/lafc-evict

Managed By

Soroush Vahidi

See all datasets managed by Soroush Vahidi.

Contact

sv96@njit.edu

How to Cite

LAFC-Evict: Learning-Augmented Cache Eviction Dataset was accessed on DATE from https://registry.opendata.aws/lafc-evict. Please cite both the dataset release and, if discussing methodology or results, the associated paper/preprint. Dataset: Vahidi, S. "LAFC-Evict: Learning-Augmented Cache Eviction Dataset (v0.3)." Hugging Face, 2026. https://huggingface.co/datasets/SoroushVahidi/lafc-evict Paper (public preprint, no peer-reviewed venue claimed): Vahidi, S. "Decision-aligned eviction-value prediction for robust learning-augmented caching." SSRN abstract ID 6636732, 2026. https://ssrn.com/abstract=6636732

Usage Examples

Tutorials

• Get To Know A Dataset: LAFC-Evict by Soroush Vahidi
  
  Amazon S3

Tools & Applications

• LAFC-Evict dataset-generation pipeline (Augmented-caching) by Soroush Vahidi