chenchiwei / tradaboost

Transfer learning algorithm TrAdaboost,coded by python
MIT License
119 stars 75 forks source link

beta算的不对? #16

Open yiyiyi4321 opened 4 years ago

yiyiyi4321 commented 4 years ago

应该是log(n) 而不是 log(n/N) ??

homily707 commented 4 years ago

对啊,我也发现这个问题,这整个代码都跑不通呀

Futuri1 commented 1 year ago

beta是这样算的,看一下论文就可以确定了,出错是因为如果目标域的少量带标签样本数量n小于迭代次数N,log会是负数,beta计算时需要对这个log开根号,所以对负数开根号会报错