[英]The gamma and cost parameter of SVM
everybody, here is a weird phenomenon when I was using libSVM to make some predictions. 大家,这是一个奇怪的现象,当我使用libSVM做一些预测时。
When I set no parameters of SVM, I will get a 99.9% performance on the testing set. 当我没有设置SVM参数时,我将在测试集上获得99.9%的性能。 While, if I set parameters '-c 10 -g 5', I will get about 33% precision on the testing set.
然而,如果我设置参数'-c 10 -g 5',我将在测试集上获得大约33%的精度。
By the way, the SVM toolkit I am using is LibSVM. 顺便说一下,我使用的SVM工具包是LibSVM。
I wonder if there is something wrong with data set. 我想知道数据集是否有问题。 And I could not figure out which result is more convincing.
我无法弄清楚哪个结果更有说服力。
You just happen to have a problem for which the default values for C
and gamma
work well (1 and 1/num_features, respectively). 你碰巧遇到了一个问题,
C
和gamma
的默认值都能正常工作(分别为1和1 / num_features)。
gamma=5
is significantly larger than the default value. gamma=5
明显大于默认值。 It is perfectly plausible for gamma=5
to induce very poor results, when the default value is close to optimal. 当默认值接近最优时,
gamma=5
导致非常差的结果是完全合理的。 The combination of large gamma
and large C
is a perfect recipe for overfitting (eg high training set performance and low test set performance). 大
gamma
和大C
的组合是过度拟合的完美配方(例如,高训练集性能和低测试集性能)。
声明:本站的技术帖子网页,遵循CC BY-SA 4.0协议,如果您需要转载,请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.