Experiments with the NNMFPACK library: influence of beta parameter in the NNMF approximation error

Autores UPV
Año
CONGRESO Experiments with the NNMFPACK library: influence of beta parameter in the NNMF approximation error

Abstract

NnmfPack is a numerical library designed to compute efficiently the Non-Negative Matrix Factorization (NNMF). It has been conceived for shared memory heterogeneous parallel systems, and it supports, from its conception, both conventional multi-core processors and many-core coprocessors. NnmfPack offers different algorithms which allow to handle different metrics options such as beta-divergence or Frobenius norm. In real applications the choice of the beta parameter presents a problem for the users of the NNMF that must decide which value to use to obtain the best approximation. In this paper, the influence of the parameter beta in the NNMF approximation error for different problems is empirically evaluated. Different datasets have been used in order to analyze and evaluate the dependency between the quality of the approximation provided by NnmfPack and the value of the beta parameter used.