摘 要:本文通过对L2正则化逻辑回归进行分析,使用随机梯度下降(SGD)和限制内存拟牛顿法(L-BFGS)来求解回归参数使得条件对数似然函数最大。在手写数字图像数据集USPS-N和HTML网页数据集上的两分类结果表明,随机梯度下降求解方法在两数据集上有较高的测试错误率。因此,在设计L2正则化逻辑回归求解方法时,可使用限制内存拟牛顿法作为缺省求解方法。
关键词:逻辑回归,随机梯度下降法,限制内存拟牛顿法
作者介绍:
黄雄鹏(1994-),男,侗族,贵州余庆人。研究方向:计算机科学与技术。
中图分类号:TP302.2 文献标识码:A 文章编号:2096-4706(2018)03-0000-03
Design of Logical RegressionSolving Based on L2 Regularization
HUANG Xiongpeng
(School of InternationalEducation,Guizhou Normal University,Guiyang 550001,China)
Abstract:By analyzing the L2 regularized logistic regression,this paperuses random gradient descent (SGD) and limited memory quasi Newton method (L-BFGS) to solve the regression parameter to make the maximum of theconditional log likelihood function. The two classification results on thehandwritten digital image data set USPS-N and the HTML web data set show thatthe random gradient descent method has a higher test error rate on the two dataset. Therefore,when designing L2 regularized logisticregression method,the restricted memory quasi Newtonmethod can be used as the default solution.
Keywords:logical regression;stochasticgradient descent method;restricted memory quasi Newtonmethod
参考文献:
[1] JANOCH A,KARAYEV S,YangqingJia,et al.A category-level 3-d object dataset: putting the kinect to work[C],2011-06-15,S.l.:s.n.,2011:1168-1174.
[2] PEDRO HTC,COIMBRA CFM. Assessment of forecasting techniques for solar power production with noexogenous inputs [J].Solar Energy,2012,86(7):2017-2028.
[3] Ding N,VishwanathanSVN.t-Logistic regression [C].Advances in Neural Information Processing Systems,2010: 514-522.
[4] Schmidt M.minFunc:unconstrained differentiable multivariate optimization in Matlab[J].Software available at http://www.cs.ubc.ca/~schmidtm/Software/minFunc.html,2005.