Stochastic Local Search: Foundations & Applications
Stochastic Local Search: Foundations & Applications
Short communication: Bootstrap quantile estimation via importance resampling
Computational Statistics & Data Analysis
A fast procedure for calculating importance weights in bootstrap sampling
Computational Statistics & Data Analysis
Hi-index | 0.00 |
Bootstrap resampling is starting to be frequently applied to contingency tables analysis of Genome-Wide SNP data, to cope with the bias in genetic effect estimates, the large number of false positive associations and the instability of the lists of SNPs associated with a disease. The bootstrap procedure, however, increases the computational complexity by a factor B, where B is the number of bootstrap samples. In this paper, we study the problem of minimizing time when applying bootstrap to contingency tables analysis and propose two levels of optimization of the procedure. The first level of optimization is based on an alternative representation of bootstrap replicates, bootstrap histograms, which is exploited to avoid unnecessary computations for repeated subjects in each bootstrap replicate. The second level of optimization is based on an ad-hoc data structure, the bootstrap tree, exploited for reusing computations on sets of subjects which are in common across more than one bootstrap replicate. The problem of finding the best bootstrap tree given a set of bootstrap replicates is tackled with best improvement local search. Different constructive procedures and local search operators are proposed to solve it. The two proposed levels of optimization are tested on a real Genome-Wide SNP dataset and both are proven to significantly decrease computation time.