<ul><li>Nearest neighbor (NN) methods have become popular for matrix completion due to their strong empirical performance and recent theoretical guarantees.</li><li>The paper introduces N$^2$, a unified Python package and testbed that consolidates a range of NN-based methods for rapid experimentation and benchmarking.</li><li>A new NN variant introduced in the framework achieves state-of-the-art results in various settings, outperforming classical methods in real-world scenarios.</li><li>The release includes a benchmark suite of real-world datasets to test matrix completion methods beyond synthetic scenarios such as healthcare, recommender systems, causal inference, and LLM evaluation.</li></ul>

N$^2$: A Unified Python Package and Test Bench for Nearest Neighbor-Based Matrix Completion

Discover more