Comparison-of-Bayesian-and-fisher

  训练错误率和交叉验证错误率相等，在样本比较大时，这个结果是可以预期的；训练错误率一般低于测试错误率，但是当样本数据比较少时，实验也出现了意外，样本多的那组测试错误率比样本少的训练错误率还要小；在本实验中，同组数据的交叉验证错误率比独立测试错误率高，这个反常现象是因为样本的原因所致，交叉验证的样本小，而独立测试时所用训练样本数目大，因而出现这种情况。分类线上，fisher准则是一条直线，而贝叶斯分类器实际上是一个类似椭圆的封闭曲线；很明显，贝叶斯分类器比fisher分类器要好。(Training error rate and cross- validation error rates are equal, the larger the sample , this result is to be expected training error rate is generally lower than the test error rate , but when comparing the sample data came from the experiment there was an accident, that more samples group test error rate less than the training sample error rate is smaller In this experiment, the same set of data cross-validation error rate than independent test error rate, this anomaly is because the sample of reasons , cross-validation sample is small,And independent testing large number of training samples used , resulting in this situation.Classification online , fisher criterion is a straight line , while the Bayesian classifier is actually a closed curve similar to elliptical It is clear that the Bayesian classifier is better than the fisher classifier .)