【摘要】 不论对分类问题还是回归问题,在构造实际可行的寻找决策函数f(x)的学习算法时,首先要有一个评价f(x)好坏的标准。而评价一个决策函数的性能时,一般是利用样本集估计其在检验集上推断时发生的错误率。给出了几个错误率估计算法,并详细分析了各估计函数的优缺点,最后的实验结果给出了从k-折交叉验证、RM-bounds和εα-estimator函数中预测出的测试错误率,进一步说明了不同的数据集可以选择不同的风险评估算法来预测出所选模型的最优参数。更多还原

【Abstract】 In order to select a good hypothesis(or model) from a collection of possible models,one has to assess the generalization performance of the hypothesis which returned by a learner that is bound to use some particular model.Several methods for estimating the generalization error of the hypotheses with least test error in the model are intro-duced in this paper,and the advantages and disadvantages of the error estimators are also analyzed in detail.The experi-mental results show different value obtained from cross-validation,RM-bounds and εα-estimator algorithms.The results also show that different problems can choose variant error estimation functions to predict the optimal parameters in the selected model.更多还原