A vectorized distributed parallel TMCMC random sampling algorithm

Abstract

The invention relates to a vectorized distributed parallel TMCMC random sampling algorithm, which comprises the following steps of: vectorizing an objective function through an element operation modeand a matrix operation mode to obtain a vectorized objective function, namely a vectorized posterior probability density function; building a distributed computing platform based on an MATLAB (MatrixLaboratory) Distributive ComputingServer tool box, and building a distributed computing platform based on the MATLAB Distributive ComputingServer tool box; And running a vectorized distributed parallel TMCMC algorithm on the distributed computing platform so as to solve a target function. According to the method, on the premise that the calculation precision is guaranteed, the calculation efficiency is greatly improved, and the calculation time is shortened.

Description

technical field [0001] The invention belongs to the random sampling algorithm technology in the Bayesian method, in particular to a vectorized distributed parallel TMCMC random sampling algorithm. Background technique [0002] When using Bayesian theory for statistical inference, theoretically speaking, for any prior distribution, it is only necessary to calculate the characteristics of the required posterior distribution according to Bayesian theory, such as the moment of the posterior distribution (the posterior mean , posterior variance), posterior probability density function, etc.; its essence is to calculate the high-dimensional integral of the function involved in the posterior distribution. However, in practical applications, the monitoring data is often scarce, and the posterior distribution of unknown parameters is often a high-dimensional, complex and uncommon distribution, which is very difficult to calculate. The random sampling method can break through this ex...

AI Technical Summary

Problems solved by technology

[0004] However, the existing TMCMC random sampling algorithm mainly has the following problems: First, because the posterior probability density function often has high dimensions and complex forms, especially the calculation cost of the likelihood function increases significantly with the increase of the number of sampling points and the sampling time, so the calculation The speed is slow; secondly, because TMCMC itself is based on the characteristics of the MCMC algorithm, and sampling is performed in stages, one operation may need to sample several or even dozens of stages, which brings more computing costs, especially when encountering large structures. The calculation time is even more astonishing, and it often takes tens of hours or even hundreds of hours for a correction, which greatly limits the application of the proposed method and most Bayesian sampling correction methods

Images