A GPU-accelerated frequent itemsets mining method based on cuda framework

Abstract

The invention belongs to the technical field of computer application, and provides a frequent item set mining algorithm based on GPU acceleration of a CUDA. The frequent item set mining algorithm based on the GPU acceleration of the CUDA adopts graph connection and dynamic queue modes. The method fully combines the advantages of an Apriori algorithm and an Eclat algorithm and converts a logic complex task for generating candidate item sets into a calculation intensive task to adapt to a calculation method of the CUDA, and the limitation of a global memory of a GPU is reasonably overcome through the dynamic queue mode, for example, after a discrete dataset is converted into a vertical bitmap, the situation is avoided that the required memory exceeds the limitation of the global memory of the GPU. It is proved through experiments that the method has the advantages that when various types of large-scale datasets are processed, the acceleration performance exceeds that of a serial algorithm, the processing capability is significantly improved, and the extracted frequent item sets are accurate and reliable. In the actual engineering application, the frequent item set mining algorithm has the advantages that other algorithms cannot replace.

Description

technical field [0001] The invention belongs to the field of computer application technology, and relates to a frequent item set mining method in an association rule mining algorithm. In particular, it relates to the method of using GPU to accelerate the algorithm of applying Apriori algorithm to frequent itemset mining on large-scale data sets. Background technique [0002] As the main step of association rule mining algorithm, frequent itemset mining can be used to derive association rules. It plays a huge role in mining the association between items and is widely used in financial forecasting, medical diagnosis and business recommendation. The purpose of frequent itemset mining is to mine potential combinations in the data set, that is, frequent items. The evaluation standard of frequent items is all itemsets whose support degree is greater than the minimum support degree threshold. [0003] The problem of this algorithm is described as follows: I={i1, i2,...,im} represe...

AI Technical Summary

Problems solved by technology

[0005] Although the above-mentioned Apriori algorithm can be accelerated in parallel through the CUDA framework, when there are many frequent itemsets or the minimum support threshold is low, the algorithm still needs to generate a large number of candidate item sets and complete a large number of candidate itemsets by repeatedly scanning the database. The support calculation of itemsets, and the IO of frequent access to external memory will lead to low efficiency of the algorithm

Although the Eclat algorithm can avoid repeated scanning of the database to complete the calculation of support, it will generate a large number of candidate item sets without combining the priori nature of Apriori, and for discrete data sets, the bitmap generated by it may require more memory. Will exceed the limit of GPU global memory

At the same time, neither of them can apply the step of candidate generation, which is a complex logic and interdependent data, to the CUDA framework

Images