Redundancy-aware sampling for Boolean matrix factorization
Description
In the associated paper, we revisit a recently published method of data reduction for Boolean matrix factorization and demonstrate that its row-selection criterion is flawed: reversing the selection logic, replacing it by simple random sampling, or by a criterion based solely on the number of 1s in a row, all yield comparable or even better results. We further propose a new row-sampling method, based on a theoretically justified score reflecting guaranteed coverage combined with randomized tie-breaking, and show experimentally that it outperforms both the examined method and the baseline approaches, especially on data with low row redundancy. This data package contains supplementary coverage and runtime tables together with folders of PDF charts that complement the findings presented in the paper.
Files
Institutions
- Univerzita Palackeho v Olomouci Katedra informatikyOlomoucký, Olomouc