| 摘要: |
| 目的 现有的参数优化方法普遍存在时间成本较大、内存占用较大、难以解决高维数据情况、难以找到全局
最优解等问题,DYCORS 算法可以在节约时间成本和内存的前提下,对高维数据问题也能找到全局最优解,故针对
现有参数优化方法存在的问题,提出了针对 OVA-SVM 模型参数分块优化的 YDYCORS 算法。 方法 OVA-SVM 的参
数中对模型影响较大的有惩罚参数 C、核函数类型 k、RBF 核函数参数 γ、ploy 核函数参数 d 以及迭代终止参数 t,由
于同时调节 5 个参数计算量较大,难以找到最优解,而 DYCORS 算法可以减少迭代次数,对于高维数据问题也同样
适用,在 DYCORS 算法的基础上进行参数分块调节:先调节影响最大的参数 C、k、γ,再固定最优参数 C、k、γ,调节
剩余参数中影响较大的参数 d 和 t,最后同时调节已获得的 5 个最优参数,如此对参数进行分块调节,提升参数优
化的效果。 结果 通过 MNIST 和 IRIS 两个数据集上的实验结果对比可以发现:运用 YDYCORS 算法对 OVA-SVM 参
数进行分块调节后,能得到与手动调参和直接用 DYCORS 同时调节 5 个参数更高的模型准确率,从而也能进一步
提升模型性能。 结论 最终实验结果表明:DYCORS 算法能有效解决 OVA-SVM 参数优化中时间成本较大、内存占
用较大、难以解决高维数据、难以找到全局最优解等问题,尤其是改进后的 YDYCORS 算法能进一步提升 OVASVM 的模型准确率,获得较佳的模型效果。 |
| 关键词: 超参数优化 支持向量机 DYCORS 算法 |
| DOI： |
| 分类号: |
| 基金项目: |
|
| Optimization and Application of OVA-SVM Parameters Based on DYCORS Algorithm |
|
YU Chenxi, YIN Yanli
|
|
School of Mathematical Sciences, Chongqing Normal University, Chongqing 401331, China
|
| Abstract: |
| Objective The existing parameter optimization methods generally have problems such as large time cost large
memory occupation difficulty in solving high-dimensional data and difficulty in finding global optimal solutions. The
DYCORS algorithm can find the global optimal solution for high-dimensional data problems even with the saving of time
cost and memory. Therefore in view of the problems existing in the existing parameter optimization methods the
YDYCORS algorithm for block optimization of OVA-SVM model parameters was proposed. Methods Among the
parameters of OVA-SVM the penalty parameter C the kernel type k the RBF kernel function parameter γ the ploy
kernel function parameter d and the iteration termination parameter t have a greater impact on the model. Due to the
large computational effort of adjusting five parameters simultaneously it is difficult to find the optimal solution. The
DYCORS algorithm can also be applied to high-dimensional data problems by reducing the number of iterations and then
the parameters were adjusted in blocks based on the DYCORS algorithm. The most influential parameters C k and γ
were adjusted first then the optimal parameters C k and γ were fixed the more influential parameters d and t among
the remaining parameters were adjusted and finally the five optimal parameters that had been obtained were adjusted simultaneously so that the parameters were adjusted in blocks to improve the effect of parameter optimization.
Results Through the comparison of the experimental results on MNIST and IRIS data sets it can be found that after using
the YDYCORS algorithm to adjust the parameters of OVA-SVM in blocks the model accuracy can be higher than the
accuracies of manual parameter adjustment and directly using DYCORS to adjust the five parameters at the same time
which can also further improve its model performance. Conclusion The final experimental results show that DYCORS
algorithm can effectively solve the problems of OVA-SVM parameter optimization such as high time cost large memory
occupation difficulty in solving high-dimensional data problems and difficulty in finding the global optimal solution. In
particular the improved YDYCORS algorithm can further improve the accuracy of the OVA-SVM model and obtain a
better model effect. |
| Key words: hyperparametric optimization support vector machine DYCORS algorithm |
