Spark-based support vector machine parameter optimization parallel particle swarm optimization method

Abstract

The invention provides a spark-based support vector machine parameter optimization parallel particle swarm optimization method which is applied to machine learning model training. The initialization information of all nodes is converted into information of an RDD form and is stored. The cross validation of node particles of different address coordinate parameters is carried out, and the node particles are mapped as an accurate rate of cross validation and the individual extremums of particles. The individual extremums of all particles are used to find a global extremum. A node particle is updated in each RDD according to the global extremum. Whether the global extremum reaches a target accurate rate or a number of iterations reaches a limit or not is judged, an optimization process is ended if so, otherwise the parallel cross validation of the node particles is repeated. The running speed of the optimization process is fast, the search range is large, and the searching of global optimum by the particle swarm algorithm is accurate and rapid.

Description

technical field [0001] The invention relates to a parallel particle swarm algorithm, and is especially suitable for a parallel particle swarm optimization method for parameter optimization of a spark-based support vector machine used in machine learning model training. Background technique [0002] A support vector machine is a machine learning method. Firstly, the optimal model parameters are obtained by cross-validating the training data set, then the optimal model parameters found in the previous step are used to train the prediction model, and finally the unknown data is predicted using the trained prediction model. [0003] The grid algorithm has already realized the optimization of parameters, and the grid algorithm has the limitation of local optimum, while the global optimization ability of the particle swarm algorithm solves the local optimum problem very well. Particle swarm optimization algorithm is a biological evolutionary algorithm that imitates the foraging o...

AI Technical Summary

Problems solved by technology

PSO involves multiple iterations, the calculation is more complex, and the single-machine implementation takes longer

Especially in the case of a large amount of data and a large number of iterations, the efficiency of the PSO stand-alone algorithm is often unacceptable.

Images