Multi-GPU density peak value clustering method based on local sensitive Hashing

Abstract

The invention belongs to the field of data mining, and provides a multi-GPU density peak value clustering method based on local sensitive Hashing. The method aims to solve the problems that when an existing density peak value clustering method is used for solving the problem of large and high-dimension data clustering, calculation complexity is too high, and consumed time is too long. The multi-GPU density peak value clustering method includes the four steps of calculation of a distance matrix, calculation of local density, calculation of a distance delta and calculation of a clustering centerand class cluster assigning. The method is characterized in that original data is divided through local sensitive Hashing, and read-write speed is increased by the utilization of a shared memory. Inthe method, local sensitive Hashing is sufficiently utilized for dividing similar data into the same GPU, so that unnecessary distance calculation is reduced. Multiple Hash functions are designed to divide the data through or operation, the situation that the similar data is mapped into different GPUs is reduced, the multi-GPU statistics method is realized based on a message passing interface (MPI), and the speed of calculating parameters is increased.

Description

technical field [0001] The invention belongs to the field of data mining, in particular to a multi-GPU (Graphic Processing Unit, graphics processor) density peak clustering method based on local sensitive hashing. Background technique [0002] Clustering is an unsupervised classification technique, the purpose of which is to divide unlabeled data sets into limited categories or clusters according to the similarity between the data, and finally make the data similarity within the group large and the difference between the groups large. Because clustering can find hidden patterns in data sets, it has been widely used in many scientific researches such as machine learning, computer vision and bioinformatics. At present, there are mainly the following clustering methods: (1) K-means and K-medoids methods, which use the center of the data point as the corresponding clustering center, can only find the shortcomings of spherical clusters, and are not suitable for non-spherical clus...

AI Technical Summary

Problems solved by technology

[0010] The present invention will aim at solving the problem that the existing density peak clustering method is too high in computational complexity and time consumption when solving large-scale and high-dimensional data set clustering problems, and provides a multi-GPU density peak based on local sensitive hashing clustering method

Images