The numerical error of a sample Mahalanobis distance (T2=y'S-1y) with sample covariance matrix S is investigated. It is found that in order to suppress the numerical error of T2, the following conditions need to be satisfied. First, the reciprocal square root of the condition number of S should be larger than the relative error of calculating floating-point real-number variables. The second proposed condition is based on the relative error of the observed sample vector y in T2. If the relative error of y is larger than the relative error of the real-number variables, the former governs the numerical error of T2. Numerical experiments are conducted to show that the numerical error of T2 can be suppressed if the two above-mentioned conditions are satisfied.
Yasuyuki KOBAYASHI
Teikyo University
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Copy
Yasuyuki KOBAYASHI, "Effects of Numerical Errors on Sample Mahalanobis Distances" in IEICE TRANSACTIONS on Information,
vol. E99-D, no. 5, pp. 1337-1344, May 2016, doi: 10.1587/transinf.2015EDP7348.
Abstract: The numerical error of a sample Mahalanobis distance (T2=y'S-1y) with sample covariance matrix S is investigated. It is found that in order to suppress the numerical error of T2, the following conditions need to be satisfied. First, the reciprocal square root of the condition number of S should be larger than the relative error of calculating floating-point real-number variables. The second proposed condition is based on the relative error of the observed sample vector y in T2. If the relative error of y is larger than the relative error of the real-number variables, the former governs the numerical error of T2. Numerical experiments are conducted to show that the numerical error of T2 can be suppressed if the two above-mentioned conditions are satisfied.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2015EDP7348/_p
Copy
@ARTICLE{e99-d_5_1337,
author={Yasuyuki KOBAYASHI, },
journal={IEICE TRANSACTIONS on Information},
title={Effects of Numerical Errors on Sample Mahalanobis Distances},
year={2016},
volume={E99-D},
number={5},
pages={1337-1344},
abstract={The numerical error of a sample Mahalanobis distance (T2=y'S-1y) with sample covariance matrix S is investigated. It is found that in order to suppress the numerical error of T2, the following conditions need to be satisfied. First, the reciprocal square root of the condition number of S should be larger than the relative error of calculating floating-point real-number variables. The second proposed condition is based on the relative error of the observed sample vector y in T2. If the relative error of y is larger than the relative error of the real-number variables, the former governs the numerical error of T2. Numerical experiments are conducted to show that the numerical error of T2 can be suppressed if the two above-mentioned conditions are satisfied.},
keywords={},
doi={10.1587/transinf.2015EDP7348},
ISSN={1745-1361},
month={May},}
Copy
TY - JOUR
TI - Effects of Numerical Errors on Sample Mahalanobis Distances
T2 - IEICE TRANSACTIONS on Information
SP - 1337
EP - 1344
AU - Yasuyuki KOBAYASHI
PY - 2016
DO - 10.1587/transinf.2015EDP7348
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E99-D
IS - 5
JA - IEICE TRANSACTIONS on Information
Y1 - May 2016
AB - The numerical error of a sample Mahalanobis distance (T2=y'S-1y) with sample covariance matrix S is investigated. It is found that in order to suppress the numerical error of T2, the following conditions need to be satisfied. First, the reciprocal square root of the condition number of S should be larger than the relative error of calculating floating-point real-number variables. The second proposed condition is based on the relative error of the observed sample vector y in T2. If the relative error of y is larger than the relative error of the real-number variables, the former governs the numerical error of T2. Numerical experiments are conducted to show that the numerical error of T2 can be suppressed if the two above-mentioned conditions are satisfied.
ER -