Markov chain Monte Carlo (MCMC)-based parallel sorting method

Abstract

The invention discloses a multi-channel multi-choice (MCMC)-based parallel sorting method. The sorting method comprises the following steps of: calculating likelihood estimation according to an initial state; calculating the posterior probability of parameters according to the likelihood estimation; performing MCMC simulation operation according to the posterior probability, and generating a novel state by taking the current state as the basis; calculating the acceptance probability according to the novel state, generating a first random number, taking the state at the next moment as a novel state when the first random number is less than the acceptance probability, otherwise keeping the current state invariable; generating a mark number of a Markov chain prepared to be exchanged in the same series of processors; and calculating the exchange probability when the Markov chain in the processor takes participates in exchange, generating a second random number, judging a comparison result of the exchange probability and the second random number, exchanging the heating parameters of the Markov chain during processing when the second random number is less than the exchange probability, otherwise avoiding exchange. According to the method, the execution time of an MC3 algorithm and an MCMC algorithm is shortened, and the communication overhead is reduced.

Description

technical field [0001] The invention relates to data classification technology, in particular to an MCMC-based parallel classification method. Background technique [0002] Aiming at the problem of data classification, there are currently many classification methods, and the single classification methods mainly include: decision tree, Bayesian, artificial neural network, K-nearest neighbor, support vector machine and classification based on association rules, etc. In addition, there are integrated learning methods for combining single classification methods, such as Bagging method and Boosting method. [0003] Among many classification methods, Bayesian classification algorithm is a kind of algorithm that uses probability and statistics knowledge to classify. When facing the classification problem of big data, the Bayesian algorithm based on statistics shows its advantages. The basic idea of ​​the Bayesian algorithm is the process of parameter post-verification probability...

AI Technical Summary

Problems solved by technology

However, in the face of a huge amount of data, the MCMC algorithm itself is very time-consuming, and the MC 3 The algorithm is more time-consuming

Images