PU Learning for Matrix Completion

Authors: Cho-Jui Hsieh, Nagarajan Natarajan, Inderjit Dhillon

ICML 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We first use synthetic data to show that our bounds are meaningful and then demonstrate the effectiveness of our algorithms in real world applications.
Researcher Affiliation Academia Cho-Jui Hsieh CJHSIEH@CS.UTEXAS.EDU Nagarajan Natarajan NAGA86@CS.UTEXAS.EDU Inderjit S. Dhillon INDERJIT@CS.UTEXAS.EDU Department of Computer Science, The University of Texas, Austin, TX 78721, USA
Pseudocode No The paper describes algorithms and optimization techniques mathematically and textually, but it does not contain a structured pseudocode or algorithm block.
Open Source Code No The paper does not provide any statement or link indicating that the source code for the described methodology is publicly available.
Open Datasets Yes We use 4 real-world datasets: 2 co-author networks ca-Gr Qc(4,158 nodes and 26,850 edges) and ca Hep Ph(11,204 nodes and 235,368 edges)...; 2 social networks Live Journal(1,770,961 nodes, |Ωtrain| = 83,663,478 and |Ωtest| = 2,055,288) and My Space(2,137,264 nodes, |Ωtrain| = 90,333,122 and |Ωtest| = 1,315,594)...We test the algorithms on two datasets: the Mushroom dataset with 8142 samples, 112 features, and 2 classes; the Segment dataset with 2310 samples, 19 features, and 7 classes.
Dataset Splits Yes in all the experiments we chose ρ from the set {1 2s, 10(1 2s), 100(1 2s), 1000(1 2s)} based on a random validation set, and use the corresponding α in the optimization problems.
Hardware Specification No The paper does not provide specific hardware details (exact GPU/CPU models, processor types, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper mentions 'eigs in MATLAB' but does not specify a version number for MATLAB or any other software libraries or dependencies with their respective versions.
Experiment Setup Yes We fix ρ = 0.9 (so that only 10% 1 s are observed). From Lemma 2, α = 0.95 is optimal. We fix k = 10, and test our algorithms with different sizes n. ... we solve the non-convex form with k = 50 for ca-Gr Qc, ca-Hep Ph and k = 100 for Live Journal and My Space. The α and λ values are chosen by a validation set.