On the Model Shrinkage Effect of Gamma Process Edge Partition Models

Authors: Iku Ohama, Issei Sato, Takuya Kida, Hiroki Arimura

NeurIPS 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We experimentally confirmed that the model shrinkage of the proposed models works well and that the IDEPM indicated state-of-the-art performance in generalization ability, link prediction accuracy, mixing efficiency, and convergence speed.
Researcher Affiliation Collaboration Iku Ohama Panasonic Corp., Japan Issei Sato The Univ. of Tokyo, Japan Takuya Kida Hiroki Arimura Hokkaido Univ., Japan
Pseudocode Yes Sampling z: Given m, similar to the Chinese restaurant process (CRP) [15], the posterior probability that zi,j,s is assigned to k is given as follows: P(zi,j,s = k | z\(ijs), m) ... Sampling m: Given z, posteriors for the φ and ψ are simulated as ... Therefore, similar to the sampler for the EPM [8], we can update m as follows: mi,j, | φ, ψ, λ δ(0) if xi,j = 0, ZTP(PK+ k=1 φi,kψj,kλk) if xi,j = 1, {mi,j,k}K+ k=1 | mi,j, , φ, ψ, λ Multinomial
Open Source Code No The paper does not provide concrete access to source code or explicitly state its release.
Open Datasets Yes The first dataset was the Enron [16] dataset, which comprises e-mails sent between 149 Enron employees. ... For larger dataset, we used the Movie Lens [17] dataset, which comprises five-point scale ratings of movies submitted by users.
Dataset Splits Yes Furthermore, all reported measurements were averaged values obtained by 10-fold cross validation.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory, or cloud instance types) used for running its experiments.
Software Dependencies No The paper does not provide specific software dependencies with version numbers.
Experiment Setup Yes We ran 600 Gibbs iterations for each model on each dataset and used the final 100 iterations to calculate the measurements. Furthermore, all reported measurements were averaged values obtained by 10-fold cross validation. ... For the purpose of fair comparison, we set hyper-hyperparameters as e0 = f0 = 0.01 throughout the experiments.