IJAPM 2017 Vol.7(3): 148-156 ISSN: 2010-362X
doi: 10.17706/ijapm.2017.7.3.148-156
doi: 10.17706/ijapm.2017.7.3.148-156
High Dimensional and Large Span Data Least Square Error: Numerical Stability and Conditionality
Vaclav Skala
Abstract—The Least Square Error (LSE) is widely used method in many engineering and computational problems solution. The LSE method is simple from the “formulation” point of view, offers a simple solution. However, it might lead to wrong or incorrect conclusions, especially if high dimensional or large span data are to be processed or if some variables have significantly smaller influence than the other does, e.g. inconvenient selection of measuring units. In this paper, we analyze influence of row and column “normalization” inspired by the Gershgorin’s theorem. The approach has been experimentally verified on a LSE application for high dimensional and large span data. The proposed approach was tested also on Hilbert’s matrix inversion for conditional number change analysis.
Index Terms—Least square error, conditionality, robustness, Gershgorin’s theorem, numerical stability.
Vaclav Skala is with Department of Computer Science and Engineering, Faculty of Applied Sciences, University of West Bohemia, CZ 306 14 Plzen, Czech Republic (email: skala@kiv.zcu.cz).
Index Terms—Least square error, conditionality, robustness, Gershgorin’s theorem, numerical stability.
Vaclav Skala is with Department of Computer Science and Engineering, Faculty of Applied Sciences, University of West Bohemia, CZ 306 14 Plzen, Czech Republic (email: skala@kiv.zcu.cz).
Cite: Vaclav Skala, "High Dimensional and Large Span Data Least Square Error: Numerical Stability and Conditionality," International Journal of Applied Physics and Mathematics vol. 7, no. 3, pp. 148-156, 2017.
General Information
ISSN: 2010-362X (Online)
Abbreviated Title: Int. J. Appl. Phys. Math.
Frequency: Quarterly
DOI: 10.17706/IJAPM
Editor-in-Chief: Prof. Haydar Akca
Abstracting/ Indexing: INSPEC(IET), CNKI, Google Scholar, EBSCO, Chemical Abstracts Services (CAS), etc.
E-mail: ijapm@iap.org
-
Dec 27, 2021 News!
IJAPM Vol 9 & Vol 10 have been indexed by Inspec [Click]
-
Jan 04, 2023 News!
The paper published in Vol 13, No 1 has received dois from Crossref
-
Jan 04, 2023 News!
IJAPM Vol 13, No 1 has been published with online version [Click]
-
Jan 04, 2023 News!
All the papers published in Vol 10-Vol 12 have received dois from Crossref
-
Sep 23, 2022 News!
IJAPM Vol 12, No 4 has been published with online version [Click]
- Read more>>