Selectivity Estimation on Synopses Based on Chebyshev Polynomials
HE Hai-jiang
Abstract
HE Hai-jiang
Abstract
Suppose we are given N values of an attribute in a relation. B (BN) numerical values are used to approximate frequency distributions of attribute. To date, two major high-accuracy synopses construction techniques have been proposed in the database literature, namely histograms and Haar wavelets. Still, these approximation techniques have not been previously compared to Chebyshev polynomials. In this paper, we conduct such a comparison the accuracy achieved with synopses based on Chebyshev polynomials to that of other techniques, when estimating frequency distributions. We demonstrate that Chebyshev synopses have three main strong points in comparision to other methods: they achieve higher accuracy as measured by the L1, L2 and L∞ error metrics; they can be built in only O(NB) time complexity; and Chebyshev synopses can be dynamically maintained. Experimental results on synthetic and real-life data clearly demonstrate the effectiveness of our method.
A significance statement is not available in the OpenAlex record.
A contribution statement is not available in the OpenAlex record.
Method details are not available in the OpenAlex metadata.
Findings are not separately available in the OpenAlex metadata.
Limitations are not available in the OpenAlex metadata.
Application details are not available in the OpenAlex metadata.
Suppose we are given N values of an attribute in a relation. B (BN) numerical values are used to approximate frequency distributions of attribute. To date, two major high-accuracy synopses construction techniques have been proposed in the database literature, namely histograms and Haar wavelets. Still, these approximation techniques have not been previously compared to Chebyshev polynomials. In this paper, we conduct such a comparison the accuracy achieved with synopses based on Chebyshev polynomials to that of other techniques, when estimating frequency distributions. We demonstrate that Chebyshev synopses have three main strong points in comparision to other methods: they achieve higher accuracy as measured by the L1, L2 and L∞ error metrics; they can be built in only O(NB) time complexity; and Chebyshev synopses can be dynamically maintained. Experimental results on synthetic and real-life data clearly demonstrate the effectiveness of our method.
Key concepts: Chebyshev filter, Chebyshev polynomials, Computer science, Chebyshev nodes, Wavelet, Histogram, Chebyshev pseudospectral method, Algorithm