I. Introduction
. 1, the AdaBoost Boosting algorithm is the loss of function of the loss index (of course, this is something which means a step forward consistent Adaboost algorithm and the loss index)
Second, details
1, the algorithm processes
2, the most important points
Error Rate: 
==
(that is, after the round, the error rate directly with the right to sub-sample error of plus re-think on it)
(1), the heavy weight of the weak classifiers to determine how
Weight 
only determined by the classification error rate e of the classifier, the range should be e [0, 0.5]
, The greater the error rate, the smaller the weight
(2) the right to determine how the sample weight
At a weight by weight of the sample on the error rate of classifiers and classifiers if the last correctly classified samples determined
(A), said first classification is correct, if the correct classification, then the far right formula will be in the form of scores, is relatively small, misclassification is a form of positive index value is relatively large, the corresponding sub-sample of the next wrong a greater weight distribution
(b), say the error rate for correctly classified samples , if the error rate is high, that is, the weight is small, this time 
, will be too large, indicating that in this case the relative classification results can not be trusted, it is worth more attention If the error rate is small, the weight is large, at this time 
, it will be relatively small, indicating that the classification results of this example relative dependability, then assign a smaller weight
For misclassified samples, if the error rate is small, heavily weighted, so at this time
，就会偏大，表示在误差率较小的分类器里边，又分错了，所以值得更多的关注，如果误差率较大，权重较小，那么此时
，就会偏小，在误差大的分类器下，分错了，情有可原，所以分配一个更小的权重。
3、强分类器
这里所有的
的和不为1
4、理论解释
由算法的推导过程可以证明，该算法的学习步骤正是一个损失函数是指数损失函数的前向分步算法的优化问题
推导:
