A Fast Non-Parametric Density Estimation Algorithm

From MaRDI portal





Non-parametric density estimation is the problem of approximating the values of probability density function, given samples from the associated distribution. Non-parametric estimation finds applications in discriminant analysis, cluster analysis, and flow calculations based on smoothed particle hydrodynamics. The usual estimators make use of kernel functions, and require on the order of \(n^2\) arithmetic operations to evaluate the density at \(n\) sample points. A sequence of special weight functions which requires an almost linear number of operations in \(n\) for the same computation is described. A cosine based estimator which is similar to kernel estimators in convergence properties, but computationally more efficient is proposed. This estimator can be viewed as a special case of the class of estimators that form a \(\delta\) sequence. Conditions for the convergence of the method and experimental verification of its accuracy as compared to kernel based estimators for practical test cases are also presented. The experiments indicate how to choose the optimal estimator as a function of the number of points while keeping the error small.











This page was built for publication: A Fast Non-Parametric Density Estimation Algorithm

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4363338)